Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

Machine Learning with Applications 15 (2024) 100517

Contents lists available at ScienceDirect

Machine Learning with Applications


journal homepage: www.elsevier.com/locate/mlwa

A comprehensive survey on machine learning applications for drilling and


blasting in surface mining
Venkat Munagala a , Srikanth Thudumu a ,∗, Irini Logothetis a , Sushil Bhandari b , Rajesh Vasa a ,
Kon Mouzakis a
a
Applied Artificial Intelligence Institute ( A2 I2 ), Deakin University, Geelong, 3216, Victoria, Australia
b MineExcellence, Bundoora, 3083, Victoria, Australia

ARTICLE INFO ABSTRACT

Keywords: Drilling and blasting operations are pivotal for productivity and safety in hard rock surface mining. These
Blasting operations are restricted due to complexities such as site-specific uncertainties, safety risks, and environmental
Drilling and economic constraints. Machine Learning (ML) is a transformative approach to tackle these complexities
Machine learning
resulting in significant cost reductions. ML applications can reduce overall blasting costs by up to 23% and
Optimization
decrease the amount of explosives by as much as 89% compared to traditional methods. This survey presents
Surface mining
a comprehensive review of how ML can be applied to optimize drill and blast designs while accounting for
its operational challenges. Our research highlights the difficulties in collecting quality site-specific data, the
complexity of interpreting this data into insightful information, the selection of ML models relating to mining
objectives, and the need for established methods to assess blast efficiency quantitatively. We provide a synthesis
of ML model development practices in drilling and blasting and demonstrate the value of ML methodologies.
Based on our survey, we present actionable recommendations for developing ML methodologies to improve
safety, reduce costs, and enhance efficiency in drilling and blasting processes. This includes establishing
standardized data schematics, multiobjective model optimization, and comprehensive evaluation metrics. These
benefits can guide mine management and engineers to adopt ML techniques and improve on-ground operational
practices. This survey aims to serve as a resource for both practitioners and researchers shaping the future
research direction in ML applications for drilling and blasting practices.

1. Introduction below the surface. The choice of mining method depends on a variety
of factors, including the type and quality of the mineral deposit, as well
Mining is the process of extracting minerals from the Earth’s sur- as economic and environmental considerations (Bhandari, 1997).
face. A mine life cycle typically consists of four stages: exploration, In surface mining, drilling and blasting are standard methods to
mine development and planning, mine operations, and mine closure. break the hard rock into desirable fragments due to their applica-
During the exploration stage, potential mining sites are surveyed and bility in various geological conditions and highly productive nature.
mapped to investigate the possible presence of mineral deposits. Mine As the primary operations in a mine operations stage, its implica-
development and planning involves the construction of infrastructure tions extend to the downstream processes (Abbaspour, Drebenstedt,
such as administration buildings, roads, power lines, and processing Badroddin, & Maghaminik, 2018). However, drilling and blasting are
plants. Mine operations are the extraction and processing of the min- cost-intensive processes accounting for 15%–60% of the total min-
eral. This includes operations such as drilling and blasting, loading, ing costs depending on geological and geographical conditions and
hauling, dumping, crushing, and screening. The final stage, mine clo- production requirements (Bilim, Çelik, & Kekeç, 2017). Therefore, it
is crucial that drill and blast designs are cost-effective and aligned
sure, involves rehabilitating the site to its pre-mining conditions to
with downstream operational requirements for the mines’ long-term
mitigate the environmental impacts of mining and ensure the long-term
viability.
sustainability of the area. Two common methods of mining are surface
Drill and blast design development is a complicated process involv-
and underground. Surface mining is used when the mineral deposits are
ing a multitude of controllable and uncontrollable factors. Controllable
near the surface, while underground mining is for mineral deposits deep

∗ Corresponding author.
E-mail addresses: t.munagala@deakin.edu.au (V. Munagala), srikanth.thudumu@deakin.edu.au (S. Thudumu), rena.logothetis@deakin.edu.au
(I. Logothetis), sushil@mineexcellence.com (S. Bhandari), rajesh.vasa@deakin.edu.au (R. Vasa), kon.mouzakis@deakin.edu.au (K. Mouzakis).

https://doi.org/10.1016/j.mlwa.2023.100517
Received 22 August 2023; Received in revised form 19 November 2023; Accepted 26 November 2023
Available online 11 December 2023
2666-8270/© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

factors can be manipulated by mine engineers, such as hole character- 2. Machine learning workflow
istics and the amount of explosives used. In contrast, uncontrollable
factors can be sporadic and difficult to predict (Bhandari, 1997), such ML is a subset of artificial intelligence that has emerged as a pivotal
as geological and meteorological conditions. The limitations and uncer- technology with transformative potential in diverse sectors, including
tainties associated with these factors, coupled with the explosive nature the mining industry. Central to the concept of ML are algorithms
of blasting, result in significant environmental effects such as ground that facilitate learning from data, thus enabling systems to identify
vibration, air overpressure, flyrock, dust, and fumes (Abbaspour et al., patterns and make predictions or decisions. This learning often out-
2018). Therefore, along with meeting the production goals of the mine, paces human capacity in terms of speed and accuracy. ML models can
it is also essential to comply with local environmental regulations to be broadly classified into four types: supervised learning, unsupervised
safeguard the environment. learning, semi-supervised learning, and reinforcement learning. Each type
has its own unique strengths, weaknesses, and uses, and the choice of
To develop drill and blast designs, mine engineers define the site-
model is highly dependent on the nature and format of the available
specific constants (Cunningham, 1983) by using empirical equations
data (Jordan & Mitchell, 2015; Sun, Cao, Zhu, & Zhao, 2019).
and experimental (trial & error) methods. This can be time-consuming
Supervised learning is often used when we have labeled data. The
and expensive. Moreover, the variables used in empirical equations
algorithm learns a mapping function from given inputs to outputs and
limit their applicability to specific sites. Recent technological advance- enables the model to make predictions when new input is provided. Re-
ments, such as the adoption of digital and automation technologies, gression and classification models are common examples of supervised
have revolutionized the data collection process at mining sites. As a learning (Kotsiantis, Zaharakis, Pintelas, et al., 2007; Stulp & Sigaud,
result, the volume and variety of data available to mine engineers have 2015). In contrast, unsupervised learning models work with unlabeled
increased significantly and can be leveraged to inform and optimize data, focusing on exploratory data analysis to uncover underlying
drill and blast designs. However, this has created challenges in terms patterns. This type of learning is particularly useful for clustering and
of data handling and processing (Asr, Kakaie, Ataei, & Mohammadi, associative tasks such as customer segmentation in marketing, helping
2019; Bołoz & Biały, 2020; Rogers et al., 2019) due to its scale and to identify distinct characteristics within data segments (Ghahramani,
complexity. 2003; Längkvist, Karlsson, & Loutfi, 2014). Semi-supervised learning
To tackle the challenges posed by big data, researchers have begun combines elements of both supervised and unsupervised learning, uti-
to harness ML techniques for the optimization of drill and blast designs, lizing a mix of labeled and unlabeled data. This approach effectively
utilizing strategies from heuristics to hybrid methods. Recent studies navigates the challenges of needing structured labeled data, making
have demonstrated that ML techniques can reduce overall costs associ- it versatile for various training scenarios (Van Engelen & Hoos, 2020;
ated with drilling and blasting. Bakhtavar, Sadiq, and Hewage (2021) Zhu, 2005). Lastly, reinforcement learning involves an agent learning
shows a 23% cost reduction in blasting operations through the applica- to make optimal decisions through trial and error, receiving rewards or
tion of ML. Furthermore, Bayat, Monjezi, Mehrdanesh, and Khandelwal penalties for actions taken. It is commonly used in fields like robotics,
(2022) and Bayat, Monjezi, Rezakhah, and Armaghani (2020) report re- gaming, and navigation where decision-making is key (Sutton & Barto,
ductions in charge per delay by 89% and 88%, respectively, through the 2018).
While a variety of ML models exist, there is no one-size-fits-all
optimization of blasting patterns using ML algorithms. Similarly, Reza-
approach to their development. Different workflows have been sug-
eineshat, Monjezi, Mehrdanesh, and Khandelwal (2020) achieved a
gested, ranging from Amershi et al.’s nine-stage model focusing on
57% reduction in charge per delay, signifying direct cost benefits in
data handling and model development to de Souza Nascimento et al.
explosive usage and associated blasting costs. These advancements
(2019)’s more compact four-stage process with additional substages.
underscore the efficiencies and importance of integrating ML models
Our study adopts the latter, as shown in Fig. 1, due to its relevance
into drilling and blasting operations, enhancing cost-effectiveness and to our specific research context.
operational optimization. Yet, this field of research is nascent, facing
issues such as data leakage and problems with reproducibility. More- • Problem understanding is domain-specific and involves defin-
over, while ML models prove effective in one site, they often require ing the goals and requirements of the model. In the context of
modifications to fit into a new environment emphasizing the need to drilling and blasting, this includes identifying the parameters that
bridge the gap between academic research and real-world industrial need optimization and establishing the metrics that the model
practices. should aim to improve, such as rock fragmentation size, ground
This survey aims to provide a wide-ranging overview of existing vibration, or cost.
ML model development practices in drilling and blasting. Neverthe- • Data handling focuses on preparing the data for building the
less, despite our rigorous approach to selecting relevant literature, ML model. This includes data collection, preprocessing, and fea-
we must acknowledge the potential for limitations in our coverage. ture engineering. In drill and blast operations, this could involve
Some relevant research might not be included due to factors such as collecting design data from sensors and equipment, cleaning the
availability, accessibility, and focus of this review. While our selection data to remove outliers or inconsistencies, and identifying the key
features or inputs of the drill and blast design that the ML model
process is thorough, it might inevitably reflect certain criteria that
will use.
might introduce potential biases.
• Model building involves selecting and training the most appro-
In this survey, we undertake a comprehensive exploration of ex-
priate ML model and evaluating its performance. In drill and blast
isting ML model development practices within the context of drilling
operations, various ML models can be considered, each with their
and blasting. We highlight current optimization techniques and their own strengths and limitations. Performance evaluation is crucial
challenges and limitations, specifically the difficulties researchers face in determining how well the model is able to predict or optimize
when developing ML models for integration into the drilling and blast- drill and blast outcomes.
ing process at mining sites. This paper begins by describing the ML • Model monitoring ensures that the ML model continues to per-
model development workflow in Section 2, followed by related sur- form well as new data is collected. It involves checking the
veys in Section 3, and the method used to conduct our survey in model’s performance over time and updating the model as nec-
Section 4. Sections 5 and 6 present a comprehensive review of meth- essary to reflect changes in the environment. For drill and blast
ods and approaches reported across literature for ML model develop- operations, this could involve continuously monitoring the actual
ment. In Section 7, we summarize the current challenges and provide outcomes of the model’s recommended drill and blast design and
recommendations for future research. updating the model if blast performance declines.

2
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Fig. 1. The four stages and the respective substages of the ML workflow (arrows denote feedback loops to the previous stage).

The ML workflow is iterative, with feedback loops between stages. to the earlier discussion on ML methods, reinforcing the importance
It is common to revise previous steps based on outcomes and insights of data quality and handling in the performance of these methods.
gained. The sequence of substages can vary based on the problem Likewise, Lawal and Kwon (2021) presents the difficulties in applying
and application domain context. In the following sections, we will advanced models in real-world mining scenarios, offering valuable rec-
explore each of these stages in more detail, focusing specifically on their ommendations to enhance model understandability and address issues
application to drill and blast operations in mining. related to data availability.
Although the aforementioned surveys offer valuable insights into
3. Related surveys various aspects of technological implementation in mining, none pro-
vide a systematic analysis of the entire ML workflow used for drill and
This section presents a comprehensive overview of recent research blast design optimization. Our study aims to analyze the literature in
on the ML models and techniques used for drilling and blasting opti- the context of this ML workflow, focusing on production, environmen-
mization in surface mining. We discuss each study’s specific focus and tal safety, and rock stability aspects of drilling and blasting. We aim to
contributions, providing insights into the current state of the field and answer the following research questions:
identifying the knowledge gap that our survey aims to address.
A holistic view of the technology landscape in mining is offered 1. What data sources and data types are used for ML models in
by various previous studies. For instance, Ali and Frimpong (2020) drilling and blasting?
explores a broad range of technological implementations across differ- 2. How is the drill and blast design data preprocessed to train the
ent mining stages. Their focus on the interconnections among mineral ML models?
exploration, mine planning, machine operations, drilling and blast- 3. What state-of-the-art ML algorithms and model development
ing, and mineral processing offers an extensive perspective. In paral- methods are applied for optimizing drill and blast design?
lel, Shahmoradi, Talebi, Roghanchi, and Hassanalian (2020) highlights 4. How are the performance of ML models evaluated in the context
how drone technology can significantly enhance data collection and of drilling and blasting?
analysis in mining operations. The application areas range from geolog- 5. What are the key challenges and potential solutions in the
ical mapping and ore control to post-blast measurements and monitor- practical application of ML models for drilling and blasting
ing, thus contributing to improved operational efficiency. While these optimization?
studies provide valuable insights, a systematic comparison reveals a gap
in the granular analysis of ML methodologies employed in drill and Through these questions, this study aims to thoroughly review the
blast design optimization. ML techniques used for drill and blast design optimization, bridging the
Specific studies focusing on the drilling and blasting stage give existing gap in the literature.
in-depth overviews and analyses of current research methodologies,
including ML and empirical models. Van der Walt and Spiteri (2020) 4. Survey methodology
examines various methods used for predicting and measuring flyrock,
spanning areas like artificial intelligence, rock engineering, empirical The survey aims to conduct a comprehensive analysis of research
and statistical analysis, and ballistics. In contrast, Yan, Hou, and Fei that utilizes ML models to optimize drill and blast design parameters in
(2020) scrutinizes different ground vibration prediction methodologies, surface mining. Our focus is on studies examining rock fragmentation,
including empirical, artificial intelligence, multiple regression, and sim- rock throw, ground vibration, flyrock, air overpressure, backbreak, and
ulation methods, and further extends to sensitivity analyses and model toe formation. For simplicity and clarity in our analysis, we classified
evaluation metrics. Dumakor-Dupey, Arya, and Jha (2021) and Al- these study objectives into three broad areas: production, environment,
Bakri and Sazid (2021) present a comprehensive overview of predictive and rock stability. The production category encompasses factors related
and optimization models for various post-blast impacts, including rock to the operational efficiency of the drilling and blasting process. The
fragmentation, ground vibration, flyrock, airblast, and backbreak. Re- environment category comprises considerations surrounding the envi-
cently, Abd Elwahab, Topal, and Jang (2023) carried out a meticulous ronmental impact of blasting. Lastly, the rock stability category includes
study of ML applications in surface mine blasting, recognizing hy- factors influencing the stability of the mining site and the safety of
brid/ensemble ML methods as more effective than others in 61% of personnel. The distribution of these categories is illustrated in Fig. 2.
the studied cases. This research also highlighted the frequent utilization To ensure thoroughness and rigor in our comprehensive survey,
of controllable parameters in blast design development, asserting their we adopted the methodology put forth by Kitchenham and Brereton
critical role in surface blasting predictions and optimizations. (2013). This systematic approach involved several key steps: identify-
Furthermore, some studies focus on the practical challenges of ing keywords and executing combinational queries in various
model development and deployment. Qi (2020) highlights the complex- databases, applying inclusion and exclusion criteria (refer to Sec-
ities inherent to data collection in mining environments. The intricacies tion 4.2) to filter the studies, and screening the filtered literature for
of this process necessitate robust data handling procedures to ensure the final qualitative synthesis. Through this process and the additional
the effective use of data in model development. This idea relates back snowballing technique to incorporate the most recent literature, we

3
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Fig. 2. A classification of blasting objectives.

Table 1
Keywords for search query.
Category Keywords
Drilling and blasting rock fragmentation, rock breakage, rock throw,
rock displacement, flyrock, ground vibration,
peak particle velocity, air overpressure, air
blast, noise, backbreak, toe, drilling, blasting,
drill and blast design, surface mining, open-pit
mining, open-cast mining
Machine learning prediction, optimization, minimization,
estimation, machine learning, artificial
intelligence

identified and selected a total of 208 relevant studies for our analysis.
Fig. 3 illustrates a detailed overview of this process. The goal was to
compile a set of studies that provided meaningful insight into the use
of ML in drilling and blasting optimization.

4.1. Corpus curation

The relevant literature for our survey was selected by establishing


a systematic approach to extract keywords to guide our search process.
We initially conducted a preliminary review of similar studies in the
field of drilling and blasting to determine frequently used terminology.
We then compiled a list of keywords relevant to our survey objective
and divided them into two categories: ‘‘drilling and blasting" and
‘‘machine learning", refer to Table 1. We constructed a search query
using these keywords with relevant synonyms. This query was executed
on three popular databases: ‘‘Google Scholar", ‘‘Scopus", and ‘‘Web of
Science’’. The search yielded 758 records, including journal articles,
conference papers, thesis reports, white papers, commercial reports,
books, and others. The snowballing process added 51 more studies to
our initial set of records, and each of these newly identified studies was
then subjected to the same filtering and inclusion process as the initial
set of records. To understand the evolution and advances of ML models
in drilling and blasting, we conducted a search from 2005 to the present
with a particular focus on the research conducted in the last five years.

Fig. 3. An overview of survey methodology.


4.2. Filtering and inclusion

We conducted manual filtering based on inclusion criteria (IC) and


exclusion criteria (EC), as shown below:

4
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Fig. 4. Temporal trends in the distribution of study categories.

IC1 Study is peer-reviewed; Table 2


IC2 Study pertains to drilling and blasting; Number of studies per category.
IC3 Study involves the development or comparison of ML models Category Study Count
in drilling and blasting scenarios;
Production 30
IC4 Study provides a description of the dataset used; (rock fragmentation, rock throw)
EC1 Study is not in English;
Environmental safety 145
EC2 Study is a duplicate or non-peer-reviewed;
(ground vibration, flyrock, air overpressure)
EC3 Study is unrelated to surface mining;
EC4 Study does not propose or compare an ML model; Rock stability 20
(back break, toe formation)
EC5 Study is unrelated to drilling and blasting outcome;
EC6 Study proposes a method or technique for drilling and Multiobjective 13
blasting outcome measurement.

In the initial screening, we identified and removed 397 records


that were duplicates, did not pertain to the domain of drilling and challenges and the resource-intensive nature of these areas. Thus, in
blasting in surface mining, or were not written in English. A further the following sections, we delve deeper into the three key stages of this
40 studies that were non-peer-reviewed or of low academic rigor were process: data overview and preprocessing, model selection and training,
excluded, such as technical reports, conference papers, and books. We and evaluation.
then screened abstracts and removed 142 records that were unrelated
to any one of the categories outlined in Fig. 2 or did not include any 5. Data overview and preprocessing
ML. The remaining 230 studies were thoroughly examined for quality
screening. 22 were rejected for lacking information on the dataset Data preprocessing is the foremost and crucial step in the ML work-
used for training the ML model or blast objective quantification studies flow. Its purpose is to ensure the quality of the data. The performance of
such as image analysis for rock fragmentation (Bamford, Esmaeili, & an ML model is directly impacted by the quality of the data it is trained
Schoellig, 2021). During this process, we encountered two studies with upon (Halevy, Norvig, & Pereira, 2009). Therefore, we initially focus
multiple experiments targeting different objectives. As these experi- on the variety of data used across the literature. We then review the
ments had distinct configurations, we treated each as a separate entity various preprocessing methods employed to prepare the data for model
for the purpose of our analysis. Consequently, 208 studies were selected training. Despite its importance, many neglect the data preprocessing
for the qualitative synthesis. stage as it is a manual process that can be time-consuming (Xin, Wu,
We categorized these 208 studies into three categories as illustrated Lee, Salehi, & Parameswaran, 2021).
in Fig. 2: Production, Environmental, and Rock Stability. Studies that
developed an ML model to concurrently predict multiple objectives that 5.1. Data understanding
crossed into more than one category were classified as Multiobjective.
The temporal trends of these categories can be seen in Fig. 4. A Drill and blast design is a complex process influenced by numerous
marked scarcity of studies prior to 2011 could be attributed to the factors, which are commonly classified into controllable and uncon-
then-nascent state of ML technology and its application to drilling and trollable factors (Bhandari, 1997). These factors are collected from
blasting. Notably, the rapid technological advancements since 2011 various sources that comprise text, images, audio, video, and spatial
have spurred an increased interest in this area, particularly from 2016 data types (Qi, 2020; Shahmoradi et al., 2020). However, data avail-
onwards. This trend underscores the necessity and timeliness of this ability and quality can vary greatly depending on mine characteristics,
survey and the need to identify efficient and effective approaches for capabilities, and methods employed for data collection. This diversity
developing ML models for drill and blast optimization. can cause distinct and varying datasets that were used in studies. This
Table 2 shows the distribution of the identified studies across the led us to analyze the dependent and independent features from the
different categories, taking into account the separately treated exper- selected literature for understandability. Based on this analysis, we
iments from the two multi-experiment studies. It is clear that there developed a comprehensive taxonomy of all the drilling and blasting
is a focus on minimizing the environmental impact and a notable factors; refer to Figs. 5 and 6.
shift towards sustainable practices in drilling and blasting where 70% Our analysis revealed that common controllable factors include
(145) of the studies prioritize Environmental Safety. The low count of burden, spacing, maximum charge per delay, powder factor, and stem-
studies on Rock Stability and Multiobjective suggest potential technical ming. Common uncontrollable factors are rock density, rock quality

5
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Fig. 5. Taxonomy of drill and blast uncontrollable factors: () for units of measurement or factor value, and {} for the number of studies that have used the factor. NA in ()
represents no unit of measurement or is derived from measurements that do not have measurements.

designation, p-wave velocity, rock mass rating, and blastability index. as duplicate records, outliers, or missing fields is challenging. These
Despite the fact that all literature agrees that uncontrollable factors anomalies result from the recording mechanisms employed in a mine
play a significant role in developing the drill and blast design, only 37% and must be addressed to ensure the robustness of the ML model.
(78) of studies have considered at least one uncontrollable factor in ML Out of the 208 studies that developed ML models, we identified
models. This exclusion could be because of the complexities involved in only 57% (118) studies provided information about the steps taken
measuring geological factors and the absence of an industry standard to prepare the data and 6% (12 studies) mentioned the removal of
parameter set. As a result, it is at the authors’ discretion to identify outliers (Arthur et al., 2022; Bayat et al., 2022, 2020; Faradonbeh,
factors to consider. Armaghani, Amnieh, & Mohamad, 2018; Jahed Armaghani et al., 2018;
The volume of data used in these studies was another important Ohadi, Sun, Esmaieli, & Consens, 2020; Yu, Koopialipoor, et al., 2021;
factor to consider, with most studies having trained their ML models Yu, Shi, Zhou, Chen, & Qiu, 2020; Yu, Shi, Zhou, Gou, et al., 2020;
on datasets comprising 100 or fewer records. This low volume may Yu, Shi, Zhou, et al., 2021; Zhang, Li, Gui, & Zhou, 2022; Ziggah,
be due to a lack of organized data collection mechanisms in a mine Temeng, & Arthur, 2023). This lack of attention to outlier removal
or the complexities involved in recording the data, such as geological may be due to the small dataset sizes, where removing records could
factors (Xie et al., 2021). decrease the data further. Hence, there was a trade-off between clean
data and dataset size. However, it is crucial to strike a balance between
5.2. Data preprocessing clean data and dataset size. A robust data cleaning procedure ensures
accurate, reliable, and representative data (García, Luengo, & Herrera,
The use of ML made data a first-class citizen. To develop a robust ML 2015), resulting in a reliable and generalized ML model.
model, the training data needs to be accurate, fair, and robust. Prepar- Another important step in the data preprocessing stage is dimen-
ing the data to develop the model involves cleaning, transformation, sionality reduction. It is a feature selection technique used to iden-
and dimensionality reduction. This section delves into the various data tify important features that impact the ML model’s output. The non-
preprocessing methods employed by researchers to create quality data. important features are ignored, resulting in a reduction of input fea-
A vital step in data preprocessing is data cleaning. Despite its crucial tures and, in turn, model complexity. A review of the literature re-
role in ensuring the quality of data for ML models, it is often over- vealed that 8% (17 out of 208) studies implemented various reduction
looked. Data cleaning involves processes such as eliminating duplicate strategies, such as correlation coefficient (Guo, Nguyen, Bui, & Ar-
records, handling missing fields, and identifying and rectifying outliers maghani, 2021), mean impact value (Zhongya & Xiaoguang, 2018),
in the data (Chu, Ilyas, Krishnan, & Wang, 2016; Rahm, Do, et al., linear modeling (Zhou, Armaghani, Ye, Khari, & Motahari, 2021),
2000). Neglecting data cleaning may lead to skewed or biased results, principal component analysis (Babaeian, Sereshki, Ataei, Nehring, &
undermining the reliability of the models (Aggarwal & Yu, 2001). In the Mohammadi, 2023; Hudaverdi, 2022), neighborhood component anal-
context of drilling and blasting, where data can be structured, semi- ysis (Ziggah et al., 2023), cosine similarity (Komadja et al., 2022),
structured, or unstructured, properly addressing data anomalies such random forest (Han, Jahed Armaghani, Tarinejad, Zhou, & Tahir, 2020;

6
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Fig. 6. A taxonomy of drill and blast controllable factors with measurement units and study frequency indicated: () for units or factor values and {} for number of studies.

Kadingdi et al., 2022), boruta algorithm (Chen, Asteris, et al., 2022), workflow for drill and blast design. We advocate for future research
Gradient Boosted Tree (GBT) (Peng, Zeng, Armaghani, Hasanipanah, in this field to rigorously and consistently apply data preprocessing
& Chen, 2021), and Chi-Square Automatic Interaction Detection algo- steps. This approach will ensure that ML models provide valuable and
rithm (CHAID) (Zeng et al., 2021). Some studies used the Fuzzy-Delphi trustworthy insights into drilling and blasting.
by leveraging expert knowledge (He et al., 2021; Li, Koopialipoor,
& Armaghani, 2021). For example, GBT Peng et al. (2021) excluded 6. Model selection, training, and evaluation
spacing to burden ratio and stemming length parameters as they were
found to be less important. Zeng et al. (2021) used the CHAID algorithm This section presents the different approaches researchers have used
to identify burden to spacing ratio and spacing features as less significant, to develop, train, and evaluate ML models. Firstly, we focus on how the
resulting in their omission from further analysis. data is split for training and testing. Then we talk about the problems of
Transformation is another crucial preprocessing step to ensure data the ML models and the approaches researchers have taken to develop
compatibility, especially when multiple data formats, such as numerical these models. Finally, we delve into model evaluations, their common
and categorical values, exist in the dataset. Although normalization can metrics, and major trends in evaluation methods. Table 3 provides a
improve a model’s performance and reliability, not all ML techniques summary of our survey covering the key aspects of each study, includ-
require data to be normalized. For instance, tree-based machine models ing ML models applied, dataset size, presence of uncontrollable factors
such as XGBoost (He et al., 2021) and random forest (Kumar, Mishra, & in the dataset, whether data preprocessing was included, the focus
Choudhary, 2022; Zhang, Zhou, et al., 2020; Zhou, Asteris, Armaghani, and approach of the model, and the evaluation methods employed.
& Pham, 2020) do not necessarily require normalized data. The choice Each study is categorized under the areas of production, environmental
of transformation thus depends on both the nature of the data and the safety, rock stability, and multiobjective based on the primary focus
selected model. Despite the common use of min–max normalization (refer to Section 4).
in the reviewed literature, the necessity of data normalization re-
mains unclear. For instance, studies Yang, Nikafshan Rad, Hasanipanah, 6.1. Data split
Bakhshandeh Amnieh, and Nekouie (2020) and Nguyen, Choi, Bui, and
Nguyen-Thoi (2019) employed similar model development strategies, Splitting data into training, validation, and testing datasets is crit-
yet their approaches to data normalization differed. Similarly, in pre- ical in developing and evaluating an ML model (Joseph, 2022). The
dicting air overpressure using artificial neural networks, Bui, Nguyen, training set allows the model to learn patterns, the validation set for
Le, Bui, and Do (2020) normalized the data, whereas Nguyen, Bui, Bui, tuning the model’s parameters, and the testing set to evaluate the
and Mai (2020) did not. model’s performance. This segregation is vital for assessing the model’s
Data preprocessing is important to the accuracy, reliability, and generalization capability — its ability to apply learned patterns to
generalizability of the ML model and should be emphasized in the ML unseen data, thereby mitigating overfitting and underfitting (Larsen &

7
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Table 3
Survey summary.
Model (Study) Dataset U. factor/s Preproc. Model development Model evaluation
size applied
Focus Approach Metrics Strategy

FA-GBM (Xie et al. (2021)) 111 No Yes P Hybrid MAE, RMSE, 𝑅2 Single
GA-ANFIS (Zhou, Li, Arslan, Hasanipanah, and Bakhshandeh Amnieh (2021)) 88 Yes Yes P Hybrid RMSE, 𝑅2 , VAF, MEDAE Single
FA-BGAM (Fang, Nguyen, Bui, Nguyen-Thoi, and Zhou (2021)) 136 No NA P Hybrid MAE, RMSE, 𝑅2 Single
GWO-SVM (Li, Yang, et al. (2021)) 76 Yes NA P Hybrid MSE, 𝑅2 , VAF Single
ANN (Mehrdanesh, Monjezi, Khandelwal, and Bayat (2021)) 353 Yes NA P Individual MAE, RMSE, VAF, 𝑅2 Single
SVM (Miao, Zhang, Wu, Li, Yan, and Lin (2021)) NA Yes NA P Individual Relative error Single
ANN (Rosales-Huamani, Perez-Alvarado, Rojas-Villanueva, and Castillo-Sequera (2020)) 47 Yes NA P Individual 𝑅2 Single
CSO (Huang, Asteris, Manafi Khajeh Pasha, Mohammed, and Hasanipanah (2020)) 75 Yes Yes P Individual RMSE, 𝑅2 , VAF, MABE Single
ACO-BRT (Zhang, Bui, Trung, Nguyen, and Bui (2020)) 136 No Yes P Hybrid RMSE, 𝑅2 , MAE Single
FA-ANFIS (Mojtahedi, Ebtehaj, Hasanipanah, Bonakdari, and Amnieh (2019)) 72 No NA P Hybrid 𝑅2 , RMSE, VAF, NS Single
PSO (Sayevand and Arab (2019)) 80 Yes Yes P Individual 𝑅2 , RMSE, VAF, MABE, NS Single
ANN (Dimitraki, Christaras, Marinos, Vlahavas, and Arampelos (2019)) 100 Yes Yes P Individual 𝑅2 , RMSE, MAE, MAPE, SS Single
ANN (Mehrdanesh, Monjezi, and Sayadi (2018)) 432 Yes Yes P Individual 𝑅2 , VAF, RMSE, MAE Single
Production

ICA (Sayevand, Arab, and Golzar (2018)) 80 Yes Yes P Individual 𝑅2 , VAF, RMSE Single
GPR (Gao, Karbasi, Hasanipanah, Zhang, and Guo (2018)) 72 No NA P Individual RMSE, PRMSE, MBE, MAPE, 𝑅2 Single
PSO-ANFIS (Hasanipanah, Amnieh, Arab, and Zamzam (2018)) 72 No Yes P Hybrid 𝑅2 , RMSE, VAF, MEDAE, NS Single
FIS (Shams, Monjezi, Majd, and Armaghani (2015)) 185 Yes Yes P Individual 𝑅2 , RMSE, VAF Single
ANFIS (Esmaeili, Salimi, Drebenstedt, Abbaszadeh, and Aghajani Bazzazi (2015)) 80 Yes Yes P Individual 𝑅2 , RMSE, VAF, MAPE Single
ANN (Enayatollahi, Aghajani Bazzazi, and Asadi (2014)) 70 Yes NA P Individual 𝑅2 , RMSE, SEP, NS Single
ANN (Monjezi, Mohamadi, Barati, and Khandelwal (2014)) 135 No Yes P Individual 𝑅2 , RMSE, VAF Single
SVM (Shi, Jian, Wu, Huang, and Wei (2012)) 102 Yes Yes P Individual 𝑅2 , RMSE, MAE Single
ANN (Kulatilake, Hudaverdi, and Wu (2012)) 109 Yes Yes P Individual 𝑅2 , RMSE Single
ANN (Bahrami, Monjezi, Goshtasbi, and Ghazvinian (2011)) 220 Yes NA P Individual 𝑅2 , RMSE Single
ANN (Monjezi, Amiri, Farrokhi, and Goshtasbi (2010)) 132 Yes Yes P Individual 𝑅2 , RMSE, Er Single
GA-SVM (Yu, Shi, Zhou, et al. (2021)) 205 Yes Yes P Hybrid 𝑅2 , MSE, VAF Ranking
EO-ANN (Yu, Shi, et al. (2021)) NA No Yes P Hybrid 𝑅2 , MSE, VAF Single
GWO-SVM (Yu, Shi, Zhou, Chen, et al. (2020)) 137 Yes Yes P Hybrid 𝑅2 , RMSE, MAE, VAF Single
JSO-LightGBM (Yari, He, Armaghani, Abbasi, and Mohamad (2023)) 234 No NA P Hybrid 𝑅2 , RMSE, MAE, MAPE Single
ANN (Amoako, Jha, and Zhong (2022)) 102 Yes Yes P Individual 𝑅2 , MSE Single
XGBoost (Chandrahas, Choudhary, Teja, Venkataramayya, and Prasad (2022)) 152 Yes NA P Ensemble 𝑅2 , RMSE, MAPE Single

MFO-SVM (Chen, Armaghani, et al. (2022)) 76 No NA P Hybrid RMSE, 𝑅2 , RMSE, MAE. VAF Ranking
ANFIS-SFS (Ye, Dalle, Nezami, Hasanipanah, and Armaghani (2020)) 62 Yes Yes P Hybrid 𝑅2 , RMSE, MAE, MAPE Single
BIENN (Temeng, Ziggah, and Arthur (2021)) 324 No Yes P Hybrid RMSE, MAPE, NRMSE, R, VAF Single
CFNN (Zeng, Jamei, Nait Amar, Hasanipanah, and Bayat (2022)) 62 Yes Yes P Individual 𝑅2 , RMSE, AARE Single
GA-ANFIS (Harandizadeh and Armaghani (2021)) 62 No Yes P Hybrid R, MSE, RMSE, Error Mean, Error StD Single
XGBoost (He et al. (2021)) 62 No NA P Ensemble R, MAE Single
GLMNETs-MLPNN (Nguyen, Bui, and Tran (2021)) 180 No Yes P Hybrid 𝑅2 , RMSE, MAE Single
ANN (Zhou, Armaghani, et al. (2021)) 62 Yes Yes P Individual RMSE, 𝑅2 , RMSE, VAF, MAE, a20-index Ranking
GP (Ramesh Murlidhar et al. (2021)) 125 Yes Yes P Individual RMSE, 𝑅2 , RMSE, MAE Ranking
Environmental Safety (Air OverPressure)

GA-GMDH (Gao, Alqahtani, Mubarakali, Mavaluru, et al. (2020)) 84 Yes NA P Hybrid 𝑅2 , RMSE, VAF, MAE, a20-index Single
ANN (Bui, Nguyen, et al. (2020)) 113 Yes Yes P Individual 𝑅2 , RMSE, MAE Single
GA-CA (Fang, Nguyen, Bui, and Tran (2020)) 164 No NA P Hybrid RMSE, 𝑅2 , MAE, VAF, MAPE Ranking
ANN (Nguyen, Bui, Bui, and Mai (2020)) 146 Yes NA P Individual RMSE, 𝑅2 Ranking
CA (Nguyen, Bui, Tran, et al. (2020)) 146 Yes NA P Individual RMSE, 𝑅2 , MAE Single
GA-BSS (Nguyen and Bui (2020)) 121 Yes Yes P Hybrid RMSE, 𝑅2 , VAF Single
FIS-FA (Zhou, Nekouie, Arslan, Pham, and Hasanipanah (2020)) 86 No Yes P Hybrid RMSE, 𝑅2 , VAF Single
ANN-RF (Nguyen and Bui (2019)) 114 No Yes P Hybrid 𝑅2 , RMSE, MAE Ranking
GA-ANN (Jahed Armaghani et al. (2018)) 97 No Yes P Hybrid 𝑅2 , RMSE, VAF Single
GEP (Faradonbeh, Hasanipanah, Amnieh, Armaghani, and Monjezi (2018)) 92 No NA P Individual 𝑅2 , RMSE, VAF Single
PSO (AminShokravi, Eskandar, Derakhsh, Rad, and Ghanadi (2018)) 80 Yes Yes P Hybrid 𝑅2 , MSE, VAF, MABE Single
PSO-SVM (Hasanipanah, Shahnazar, Bakhshandeh Amnieh, and Jahed Armaghani (2017)) 83 No Yes P Hybrid R, RMSE Single
ICA-ANN (Jahed Armaghani, Hasanipanah, and Tonnizam Mohamad (2016)) 70 No Yes P Hybrid 𝑅2 , RMSE, VAF Ranking
GA-ANN (Tonnizam Mohamad, Jahed Armaghani, Hasanipanah, Murlidhar, and Alel (2016)) 76 No Yes P Hybrid 𝑅2 , VAF, RMSE Ranking
ANFIS (Hasanipanah, Jahed Armaghani, Khamesi, Bakhshandeh Amnieh, and Ghoraba (2016)) 77 No Yes P Individual 𝑅2 , RMSE, VAF Ranking
ANFIS (Jahed Armaghani, Hajihassani, Sohaei, et al. (2015)) 128 No Yes P Individual 𝑅2 , RMSE, VAF Single
ICA-ANN (Jahed Armaghani, Hajihassani, Marto, Shirani Faradonbeh, and Mohamad (2015)) 95 No Yes P Hybrid 𝑅2 , RMSE Single
PSO-ANN (Hajihassani, Armaghani, Sohaei, Mohamad, and Marto (2014)) 62 Yes Yes P Hybrid R, MSE Single
SVM (Khandelwal and Kankar (2011)) 75 No NA P Individual 𝑅2 , MAPE Single
BIENN (Temeng, Ziggah, and Arthur (2020)) 171 No Yes P Hybrid R, RMSE, MAPE, NRMSE, VAF, PI Single
CWNN (Hosseini, Poormirzaee, and Hajihassani (2022)) 90 No Yes P Individual 𝑅2 , RMSE, VAF, Accuracy indices Single
ANN (Ziggah et al. (2023)) 324 No Yes P Individual RMSE, SI, R, CRM, U95, BIC Single
RBFN (Zhang et al. (2022)) 76 No Yes P Individual RMSE, MAPE, R Single
SVM-GO (Chen, Asteris, et al. (2022)) 62 Yes Yes P Hybrid 𝑅2 , RMSE, MAE, MAPE Single

(continued on next page)

Goutte, 1999). Typically, researchers use a two-way split of the data, Six studies reported a three-way split introducing a validation set
consisting of a training set and a test set. Out of 208 studies, 90% where 10% to 20% of the data is allocated (Azimi et al., 2019; Babaeian
(188) specified their data split strategies for model development and et al., 2023; Guo, Nguyen, et al., 2021; Ohadi et al., 2020; Trivedi et al.,
validation. The most common strategy was the two-way split, with 182 2015; Yari et al., 2016). The validation set serves for hyperparameter
studies adopting various ratios for the split: tuning and acts as an unbiased evaluation of the model’s fit on the
training data (Azimi et al., 2019). To further address the challenges of
overfitting or underfitting, researchers have adopted cross-validation
• 90%–10% (13 studies)
techniques. Typically k-fold cross-validation is used to divide the data
• 85%–25% (7 studies) into ‘k’ subsets and iteratively train the model on ‘k-1’ subsets while
• 80%–20% (124 studies) testing it on the remaining subset. This process is repeated ‘k’ times,
• 75%–25% (9 studies) and the results are averaged to provide a robust estimate of model
• 70%–30% (28 studies) performance (Barkhordari et al., 2022; Chen, Armaghani, et al., 2022;
• 60%–40% (1 study) Li, Yang, et al., 2021; Nguyen & Bui, 2020; Nguyen, Bui, Bui, & Mai,

8
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Table 3 (continued).
Model (Study) Dataset U. factor/s Preproc. Model development Model evaluation
size applied
Focus Approach Metrics Strategy

CWNN (Hosseini, Poormirzaee, Hajihassani, and Kalatehjari (2022)) 416 No Yes P Hybrid 𝑅2 , RMSE, VAF Ranking
DNN (Barkhordari, Armaghani, and Fakharian (2022)) 262 No NA P Ensemble MSE, 𝑅2 , COV, a20-index Ranking
GEP (Monjezi, Dehghani, Shakeri, and Mehrdanesh (2021)) 152 No NA P Individual 2
𝑅 , RMSE Single
FA-ANN (Li, Koopialipoor, and Armaghani (2021)) 113 No NA P Hybrid 2
𝑅 , RMSE Ranking
HHO-ANN (Murlidhar et al. (2021)) 152 Yes Yes P Hybrid 𝑅2 , RMSE, MAPE, VAF, RSE, a10-index Single
ELM (Jamei, Hasanipanah, Karbasi, Ahmadianfar, and Taherifar (2021)) 73 No Yes P Individual R, RMSE, MAPE, U1, U2 Ranking
WO-SVM (Nguyen, Bui, Choi, Lee, and Armaghani (2021)) 210 No Yes P Hybrid 𝑅2 , RMSE Single
GEP-CO (Dehghani, Pourzafar, et al. (2021)) 318 No Yes O Hybrid 𝑅2 , RMSE Single
SVM-GLMNET (Guo, Nguyen, et al. (2021)) l 210 No NA O Ensemble/Hybrid 𝑅2 , RMSE, MAPE, MAE, VAF Ranking
DNN-WO (Guo, Zhou, Koopialipoor, Jahed Armaghani, and Tahir (2021)) 240 No Yes O Hybrid 2
𝑅 , RMSE Ranking
GP-MCS (Ye, Koopialipoor, Zhou, Armaghani, and He (2021)) 262 No NA O Individual 𝑅2 , RMSE, VAF Ranking
GP-RFNN (Nikafshan Rad, Bakhshayeshi, Wan Jusoh, Tahir, and Foong (2020)) 70 No Yes P Hybrid 𝑅2 , RMSE Single
BO-ELM (Murlidhar et al. (2020)) 262 No Yes P Hybrid 𝑅2 , MAE, NSE, Cp, RMSE Ranking
ANN-ADHS (Hasanipanah, Keshtegar, Thai, and Troung (2020)) 82 Yes NA P Hybrid 2
𝑅 , RMSE, MAE, NSE, d Single
SVM-GWO (Armaghani, Koopialipoor, Bahri, Hasanipanah, and Tahir (2020)) 262 No NA O Hybrid 𝑅2 , RMSE Single
Environmental Safety (Flyrock)

ANN-MCS (Zhou, Aghili, et al. (2020)) 260 No NA O Individual 2


𝑅 , RMSE Ranking
BNN (Han et al. (2020)) 262 No Yes P Individual Cumulative Gain Single
ANN-PSO (Zhou, Koopialipoor, et al. (2020)) 65 No NA O Hybrid 𝑅2 Single
PSO-ANN (Koopialipoor, Fallah, Armaghani, Azizi, and Mohamad (2019)) 262 No NA P Hybrid 𝑅2 , RMSE Ranking
ANN (Nguyen, Bui, Nguyen-Thoi, Ragam, and Moayedi (2019)) 201 No Yes P Ensemble 𝑅2 , RMSE, VAF, MAPE, MAE Ranking
SVM (Rad, Hasanipanah, Rezaei, and Eghlim (2018)) 90 Yes NA P Individual 2
𝑅 , MSE, VAF, MABE Single
GEP-FA (Faradonbeh, Armaghani, et al. (2018)) 76 No Yes O Hybrid 𝑅2 , VAF, RMSE Single
PSO (Hasanipanah, Jahed Armaghani, Bakhshandeh Amnieh, Majid, and Tahir (2017)) 76 Yes Yes P Individual 𝑅2 , RMSE, NS Single
RT (Hasanipanah, Faradonbeh, Armaghani, Amnieh, and Khandelwal (2017)) 65 No NA P Individual 𝑅2 , MAE, MEDAE Single
ANN (Yari, Bagherpour, Jamali, and Shamsi (2016)) 334 Yes Yes P Individual R, RMSE Single
GP (Faradonbeh, Jahed Armaghani, and Monjezi (2016)) 262 No NA P Individual 2
𝑅 , RMSE, VAF Ranking
GEP (Faradonbeh, Armaghani, Monjezi, and Mohamad (2016)) 97 No NA P Individual 2
𝑅 , RMSE, VAF, MAE Single
ANFIS (Trivedi, Singh, and Gupta (2015)) 125 Yes Yes P Individual 𝑅2 , RMSE, MAE Single
ICA-ANN (Marto, Hajihassani, Jahed Armaghani, Tonnizam Mohamad, and Makhtar (2014)) 113 Yes NA P Hybrid 𝑅2 , RMSE, MAE Single
ANFIS (Ghasemi, Amini, Ataei, and Khalokakaei (2014)) 230 No Yes P Individual 𝑅2 , RMSE, VAF, MAPE Single
ANN (Monjezi, Mehrdanesh, Malek, and Khandelwal (2013)) 310 Yes NA P Individual RMSE, VAF, 𝑅2 , MAE, Er Single
SVM (Amini, Gholami, Monjezi, Torabi, and Zadhesh (2012)) 245 No Yes P Individual R, RMSE Single
ANN (Monjezi, Bahrami, Varjani, and Sayadi (2011)) 213 Yes NA P Individual 2
𝑅 , RMSE, MAE, Er Single
AdaBoost (Yari, Armaghani, et al. (2023)) 234 No NA P Ensemble 𝑅2 , RMSE, VAF, a10-index Single
ANFIS (Hudaverdi (2022)) 109 Yes Yes P Individual R, MAE, MAPE, NSE Single
EO-ELM (Bhatawdekar et al. (2023)) 114 Yes NA P Hybrid 2
𝑅 , RMSE, MAPE, NSE, VAF, a20-index Single
GA-ANN (Babaeian et al. (2023)) 118 Yes Yes P Hybrid 𝑅2 , RMSE, VAF, MAPE Single
HLO-SVM (Huang and Xue (2022)) 240 No NA P Hybrid 𝑅2 , RMSE, MSE, MAE Single
JSO-ANN (Wang, Hosseini, Jahed Armaghani, and Tonnizam Mohamad (2023)) 65 No Yes P Hybrid 𝑅2 , RMSE, VAF Single
SVM-WO (Ding, Jamei, Hasanipanah, Abdullah, and Le (2023)) 80 Yes Yes P Hybrid 𝑅2 , RMSE, AARE Single

RVM-HPSOGWO (Yu, Shi, Zhou, Gou, et al. (2020)) 117&50a Yes Yes P Hybrid 𝑅2 , RMSE, MAE Ranking
PSO-ANN (Peng et al. (2021)) 93 Yes Yes P Hybrid R, RMSE, VAF, a20-inex Ranking
FA-SVM (Ding, Hasanipanah, Nikafshan Rad, and Zhou (2021)) 87 No NA P Hybrid 𝑅2 , RMSE, MAE, SMAPE, LM, VAF Single
CO-ANN (Bui, Nguyen, Tran, Nguyen, and Bui (2021)) 118 No Yes P Hybrid 𝑅2 , RMSE, MAE, VAF, a20-index Single
AENN-SVM (Ke, Nguyen, Bui, and Costache (2021)) 297 Yes Yes P Hybrid 𝑅2 , RMSE Single
SVM (Zeng et al. (2021)) 166 No Yes P Individual R, MAE, Gain Ranking
PSO-CRANFIS (Zhu, Rad, and Hasanipanah (2021)) 84 No Yes P Hybrid 𝑅2 , RMSE, MAE, MAPE, VAF Ranking
GO-ELM (Yu, Koopialipoor, et al. (2021)) 166 No Yes P Hybrid 𝑅2 , RMSE, MAE, MAPE, NSE Single
FA-SVM (Chen, Hasanipanah, Nikafshan Rad, Jahed Armaghani, and Tahir (2021)) 95 Yes NA P Hybrid 𝑅2 , RMSE, WI, MAE Single
JA-XGBoost (Zhou, Qiu, Khandelwal, Zhu, and Zhang (2021)) 150 Yes Yes P Hybrid 𝑅2 , RMSE, MAE, VAF Ranking
WO-XGBoost (Qiu et al. (2021)) 150 Yes NA P Hybrid 𝑅2 , VAF, RMSE, MAE Ranking
PSO-ELM (Jahed Armaghani, Kumar, Samui, Hasanipanah, and Roy (2021)) 102 No NA P Hybrid 𝑅2 , RMSE, NSE, MAE, Cp Single
GWO-RVM (Fattahi and Hasanipanah (2021)) 95 No Yes P Hybrid R, MSE, MAPE, VAF Single
ANN (Lawal, Kwon, Hammed, and Idris (2021)) 100 Yes Yes P Hybrid 𝑅2 Single
HHO-DNN (Nguyen, Bui, Tran, Nguyen, et al. (2021)) 229 No Yes P Hybrid 𝑅2 , RMSE, MAPE, VAF Single
Environmental Safety (Ground Vibration)

GEP-MCS (Zhou, Li, Koopialipoor, Jahed Armaghani, and Thai Pham (2021)) 102 No NA O Individual 𝑅2 , RMSE, MAE, VAF Single
GEP-GO (Bayat et al. (2022)) 151 No Yes O Hybrid 𝑅2 , RMSE, RMSER, MAE Single
GA-ANFIS (Yang, Hasanipanah, Tahir, and Bui (2020)) 86 No Yes P Hybrid 𝑅2 , RMSE, VAF, MAPE, MAE Single
IM-ANN (Amiri, Hasanipanah, and Bakhshandeh Amnieh (2020)) 92 No Yes P Hybrid 𝑅2 , RMSE, MAE, MAPE Single
GMDH-NN (Mohammadi, Mikaeil, and Abdollahi-Sharif (2020)) 52 No NA C Individual Accuracy, Error, Recall, F1 Score Ranking
ICA-XGBoost (Ding, Nguyen, Bui, Zhou, and Moayedi (2020)) 136 No NA P Ensemble/Hybrid 𝑅2 , RMSE, MAE Ranking
RF (Zhang, Zhou, et al. (2020)) 102 No Yes P Individual VAF, RMSE, 𝑅2 , MAE, a20-index Ranking
HKM-SVM (Nguyen, Drebenstedt, Bui, and Bui (2020)) 185 No NA P Hybrid 𝑅2 , RMSE, VAF Single
FA-SVM (Yang, Nikafshan Rad, et al. (2020)) 90 No NA P Hybrid 𝑅2 , RMSE, MAE Single
GA-SVM (Nguyen, Choi, et al. (2019)) 125 No Yes P Hybrid 𝑅2 , RMSE, MAE Ranking
FCM-QRNN (Bui, Choi, et al. (2020)) 83 No NA P Hybrid 𝑅2 , RMSE, MAE Single
ANN-FA (Bayat et al. (2020)) 151 No Yes O Hybrid VAF, RMSE, 𝑅2 , MAE, a20-index Single
ANN-ICA (Rezaeineshat et al. (2020)) 112 Yes Yes O Hybrid 𝑅2 , RMSE, VAF Single
RF (Zhou, Asteris, et al. (2020)) 102 No Yes C Individual Accuracy Single
HHO-RF-MCS (Yu, Shi, Zhou, Chen, and Qiu (2020)) 117 Yes Yes O Hybrid 𝑅2 , RMSE, MAE Single
GEP-MCS (Mahdiyar et al. (2020)) 149 No NA O Hybrid 𝑅2 , RMSE Ranking
PSO-KNN (Bui, Jaroonpattanapong, Nguyen, Tran, and Long (2019)) 152 No NA P Hybrid 𝑅2 , RMSE, MAE Single
ANN (Nguyen, Bui, and Moayedi (2019)) 68 No NA P Individual 𝑅2 , RMSE, MAE Single
BGAM (Nguyen, Bui, Tran, and Moayedi (2019)) 79 No Yes P Ensemble 𝑅2 , RMSE, MAE Single
CA-HKM (Nguyen, Bui, Tran, and Mai (2019)) 136 No Yes P Hybrid 𝑅2 , RMSE, MAE Single
XGBoost (Nguyen, Bui, Bui, and Cuong (2019)) 146 No Yes P Ensemble 𝑅2 , RMSE Ranking
GA-ANN (Azimi, Khoshrou, and Osanloo (2019)) 70 No NA P Hybrid 𝑅2 , RMSE Single
ELM (Zhongya and Xiaoguang (2018)) 108 Yes Yes P Individual 𝑅2 Single
ICA (Behzadafshar, Mohebbi, Tehrani, Hasanipanah, and Tabrizi (2018)) 76 No Yes P Individual 𝑅2 , MBE, MAPE, RMSE, VAF Single
CART (Hasanipanah, Faradonbeh, Amnieh, Armaghani, and Monjezi (2017)) 86 No NA P Individual 𝑅2 , RMSE, NS Ranking
CO (Fouladgar, Hasanipanah, and Bakhshandeh Amnieh (2017)) 85 No Yes P Individual 𝑅2 , RMSE Single
PSO (Hasanipanah, Naderi, Kashir, Noorani, and Zeynali Aaq Qaleh (2017)) 80 No Yes P Individual 𝑅2 , RMSE, VARE, NS Single
CART (Khandelwal et al. (2017)) 51 No NA P Individual 𝑅2 , RMSE, VAF Single
ABC-ANN (Taheri, Hasanipanah, Golzar, and Majid (2017)) 89 No Yes P Hybrid 𝑅2 , RMSE, MAPE Single

(continued on next page)

9
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Table 3 (continued).
Model (Study) Dataset U. factor/s Preproc. Model development Model evaluation
size applied
Focus Approach Metrics Strategy

GEP-CO (Faradonbeh and Monjezi (2017)) 115 No NA O Hybrid 𝑅2 , RMSE, MAE Single
GEP (Shirani Faradonbeh, Jahed Armaghani, et al. (2016)) 102 No NA P Individual 𝑅2 , RMSE, VAF, MAE Single
GEP 35 Yes NA P Individual 𝑅2 , RMSE, VAF Single
(Monjezi, Baghestani, Shirani Faradonbeh, Pourghasemi Saghand, and Jahed Armaghani (2016))
ICA-ANN (Hajihassani, Jahed Armaghani, Marto, and Tonnizam Mohamad (2015)) 95 Yes NA P Hybrid 𝑅2 , RMSE Single
ANFIS (Armaghani, Momeni, Abad, and Khandelwal (2015)) 109 No Yes P Individual 𝑅2 , RMSE, VAF Single
SVM (Hasanipanah, Monjezi, Shahnazar, Armaghani, and Farazmand (2015)) 80 No NA P Individual 𝑅2 , RMSE, VAF, NSE, VARE, MAPE, MEDAE Single
Environmental Safety (Ground Vibration)

FIS (Ghasemi, Ataei, and Hashemolhosseini (2013)) 120 No NA P Individual 𝑅2 , VAF, RMSE, MAPE Single
ANN (Görgülü et al. (2013)) 36 Yes NA P Individual 𝑅2 Single
SVM (Longjun, Xibing, Ming, and Qiyue (2011)) 108 Yes NA P Individual Error ratio Single
SVM (Khandelwal (2011)) 150 No NA P Individual 𝑅2 , MAE Single
ANN (Mohamed (2009)) 149 Yes NA P Individual 𝑅2 , MSE Single
ANN (Khandelwal and Singh (2009)) 174 Yes Yes P Individual 𝑅2 , MAE Single
ANN (Singh and Singh (2005)) 215 No NA P Individual 𝑅 2 Single
FA-ANN (Shang, Nguyen, Bui, Tran, and Moayedi (2020)) 83 No NA P Hybrid 𝑅2 , RMSE, MAE, VAF Ranking
ICA (Armaghani, Hasanipanah, Amnieh, and Mohamad (2018)) 73 No Yes P Individual 2
𝑅 , RMSE Single
ANN (Jain, Agrawal, and Choudhary (2022)) 56 No NA P Individual 𝑅2 , RMSE Single
BNN (Fissha, Ikeda, Toriya, Adachi, and Kawamura (2023)) 100 No Yes P Individual 𝑅2 , RMSE, MSE Single
MARS (Komadja et al. (2022)) 1001 No Yes P Individual 𝑅2 , RMSE Single
SSO-ELM (Nguyen, Bui, and Topal (2023b)) 216 Yes Yes P Hybrid 2
𝑅 , RMSE, MAE Single
RF-GBM (Kadingdi et al. (2022)) 196 No Yes P Ensemble/Hybrid 2
𝑅 , RMSE, MAE Single
XGBoost (Hosseini, Pourmirzaee, Armaghani, and Sabri Sabri (2023)) 130 No Yes P Ensemble 𝑅2 , RMSE, MAE, VAF, Accuracy Single
GPR (Arthur et al. (2022)) 101 No Yes P Individual 𝑅2 , R, VAF, MSE Single
SVM (Temeng, Arthur, and Ziggah (2022)) 201 No Yes P Individual 𝑅2 , RMSE, MSE, Elg Single
RF-GA-ANN (Guo, Yang, Wang, and Xiang (2022)) 154 No Yes P Hybrid 𝑅2 , RMSE, Er Single
RSM-SVM (Keshtegar et al. (2023)) 90 Yes NA P Hybrid 𝑅2 , RMSE, MAE, NSE, AI Single
MRFO-SONN (Nguyen, Bui, and Topal (2023a)) 288 Yes Yes P Hybrid 2
𝑅 , RMSE, MAE, MAPE Single
XGBoost (Chandrahas et al. (2022)) 152 Yes NA P Ensemble 𝑅2 , RMSE, MAPE Single
GP (Kumar, Mishra, and Choudhary (2023)) 140 No NA P Individual 𝑅2 Single

RF (Kumar et al. (2022)) 140 Yes NA P Individual 𝑅2 , RMSE Single


PSO-RF (Dai et al. (2022)) 234 No NA P Hybrid 𝑅2 , RMSE, MAE, MAPE Ranking
GP (Sharma, Agrawal, and Choudhary (2022)) 84 No NA P Individual 𝑅2 , RMSE, MAPE, MMRE Single
FIS-ICA (Hasanipanah and Bakhshandeh Amnieh (2021) 62 No NA P Hybrid 𝑅2 , RMSE, VAF, MAPE, MAE Single
MFO-SVM (Yu, Monjezi, et al. (2021)) 85 Yes NA P Hybrid 𝑅2 , RMSE, VAF, MAE Ranking
SCA-RF (Zhou, Dai, et al. (2021)) 234 No NA P Hybrid 𝑅2 , RMSE, VAF Ranking
PSO (Eskandar, Heydari, Hasanipanah, Masir, and Derakhsh (2018)) 84 Yes Yes P Individual 𝑅2 , RMSE Single
Rock Related

PSO-ANFIS (Hasanipanah, Shahnazar, Arab, Golzar, and Amir (2017)) 80 No Yes P Hybrid 𝑅2 , RMSE, MAE Single
PSO (Ghasemi (2017)) 175 No Yes P Individual 𝑅2 , RMSE, VAF Ranking
GP (Shirani Faradonbeh, Monjezi, and Jahed Armaghani (2016)) 175 No NA P Individual 𝑅2 , RMSE, VAF Single
ANFIS (Ghasemi, Amnieh, and Bagherpour (2016)) 175 No Yes P Individual 𝑅2 , RMSE, VAF, MAPE Single
ANFIS (Esmaeili, Osanloo, Rashidinejad, Aghajani Bazzazi, and Taji (2014)) 42 No Yes P Individual 𝑅2 , RMSE, MAPE Single
SVM (Mohammadnejad, Gholami, Sereshki, and Jamshidi (2013)) 135 No NA P Individual R, RMSE Single
ANN (Monjezi, Ahmadi, Varjani, and Khandelwal (2013)) 97 Yes NA P Individual 𝑅2 , VAF, MSE Single
SVM (Khandelwal and Monjezi (2013)) 234 No NA P Individual 𝑅2 , MAE Single
ANN (Monjezi and Dehghani (2008)) 300 No NA P Individual 𝑅2 , RMSE Single
ANN-ICA (Sadeghi, Monjezi, and Jahed Armaghani (2020)) 100 No Yes O Hybrid 𝑅2 , RMSE, VAF Single
RF (Sharma, Choudhary, and Agrawal (2022)) 100 No NA P Individual 𝑅2 , RMSE Single
GP (Kumar et al. (2023)) 140 Yes NA P Individual 𝑅2 Single

ANN-ICA (Zangoei, Monjezi, Armaghani, Mehrdanesh, and Ahmadian (2022)) 172 Yes Yes O Hybrid 𝑅2 , RSME, VAF, MAE Single
ANN (Gorai, Himanshu, and Santi (2021)) 101 No Yes P Individual 𝑅2 , RMSE, MSE, FB Single
RF (Ohadi et al. (2020)) 451 Yes Yes P Individual MAE, MSE Single
ANN-FA (Asl, Monjezi, Hamidi, and Armaghani (2018)) 200 Yes NA O Hybrid 𝑅2 , RMSE, Er Single
Multiobjective

ANN-KNN (Amiri, Bakhshandeh Amnieh, Hasanipanah, and Mohammad Khanli (2016)) 75 No Yes P Hybrid 𝑅2 , RMSE, VAF Yes
ANN-ABC (Ebrahimi, Monjezi, Khalesi, and Armaghani (2016)) 34 No Yes O Hybrid 𝑅2 , RMSE, VAF Single
ANN-ACO (Saghatforoush, Monjezi, Shirani Faradonbeh, and Jahed Armaghani (2016)) 97 No Yes O Hybrid 𝑅2 , RMSE, Ea Single
PSO-ANN (Hajihassani, Jahed Armaghani, Monjezi, Mohamad, and Marto (2015)) 88 Yes Yes P Hybrid 𝑅2 , MSE Single
PSO-ANN (Armaghani, Hajihassani, Mohamad, Marto, and Noorani (2014)) 44 Yes NA P Hybrid 𝑅2 , MSE, Error Single
ANN (Sayadi, Monjezi, Talebi, and Khandelwal (2013)) 103 No NA P Individual 𝑅2 , RMSE, MRE, VAF Single
GA-ANN (Monjezi, Khoshalan, and Varjani (2011)) 195 Yes NA P Hybrid 𝑅2 , MSE, NMSE Single
ANN-GA (Monjezi, Amini Khoshalan, and Yazdian Varjani (2012)) 234 No Yes O Hybrid 𝑅2 Single
ANN (Monjezi, Bahrami, and Varjani (2010)) 250 Yes NA P Individual 𝑅2 , RMSE, MAE, Er Single

‘‘NA’’: Not Available; ‘‘P’’: Prediction; ‘‘O’’: Optimization; ‘‘C’’: Classification.


a The publication has provided two independent case studies using the same ML model.

2020; Nguyen, Bui, Choi, et al., 2021; Nguyen, Bui, & Tran, 2021; it can increase bias resulting in underfitting where the model does
Nguyen, Bui, Tran, & Moayedi, 2019; Nguyen, Bui, Tran, et al., 2020; not learn enough from the training data to make accurate predictions.
Ohadi et al., 2020; Shi et al., 2012; Xie et al., 2021; Yu, Shi, et al., Balancing these aspects is critical for the reliability and effectiveness of
2021; Yu, Shi, Zhou, Gou, et al., 2020). Other strategies reported in the the model (Arlot & Celisse, 2010; Kohavi et al., 1995; Larsen & Goutte,
literature include leave-one-out cross-validation (Amini et al., 2012), 1999; Xu & Goodacre, 2018).
which can be viewed as k-fold with ‘k’ equal to the total number of Our evaluation reveals that the choice of ‘k’ in k-fold
data points, the holdout method (Arthur et al., 2022; Temeng et al., cross-validation often appears arbitrary, where researchers are not
2022), which is essentially a simple train/test split, and generalized providing a rationale for their choice. Furthermore, our review did not
cross-validation, which uses a particular formula to compute the error identify a trend between the chosen data split strategy and the reported
estimate (Komadja et al., 2022). performance of the models. This suggests that other factors play a larger
Choosing the right split ratio is crucial in a simple train–test split or role in determining model performance, such as model architecture or
when selecting the ‘k’ value for k-fold cross-validation. A training set the specific characteristics of the dataset.
larger than the test set or a smaller ‘k’ value implies that the model has
more data to learn from. Despite reducing bias, this method can lead to 6.2. Model development
overfitting, where the model becomes too tailored to the training data
and performs poorly on unseen data. On the contrary, a larger test set Researchers have experimented with various methods and tech-
or a larger ‘k’ value might decrease the risk of overfitting. However, niques in developing ML models for blast design optimization. We

10
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Fig. 7. Year-wise distribution of studies by model category (Hybrid/Single).

identify prediction and optimization as the main focuses of these mod- Together, both charts in Figs. 7 and 8 provide a detailed view of the
els. Different ML models have been developed based on individual evolution in research paradigms in the field of drilling and blasting,
and hybrid approaches (Xie et al., 2021; Zhang, Zhou, et al., 2020). showcasing a shift from individual models to hybrid models. These
The choice of approach depends on the specific problem and avail- trends reveal a broader diversification in research approaches and a
able dataset. For simple problems where a single model is sufficient willingness to explore complex, integrative ML modeling techniques
to capture the patterns in the data, an individual approach is em- for drilling and blasting. The following sections will delve into the
ployed (Nguyen, Bui, & Moayedi, 2019). For complex problems where a specific trends, characteristics, and popular models within individual
single model may not capture all the necessary patterns or relationships and hybrid development approaches.
in the data, a hybrid approach is chosen (Li, Yang, et al., 2021).
Individual approach. For individual model development, a single ML
6.2.1. Model development focus model is developed to predict outcome/s. Researchers have used indi-
The areas of interest include prediction, optimization, and classifi- vidual models to predict single (Mehrdanesh et al., 2021) and multiple
cation. However, the majority of studies focused on the former two. blast objectives (Gorai et al., 2021). These models are trained on his-
Only two studies (Mohammadi et al., 2020; Zhou, Asteris, et al., 2020) torical drill and blast design data, enabling them to learn relationships
used classification techniques specifically to classify ground vibration and patterns within the data and use the relationship learned to predict
into different ranges. the blast outcome/s for unseen drill and blast designs. A variety of
ML models have been developed where the choice of model depends
Prediction. Predictive modeling is a technique used by ML models to on various factors, including the complexity of data, the specific blast
learn (i) relations in variables and (ii) hidden patterns from historical objectives, and the computational resources available. These models
data, and (iii) predict the outcome/s of unseen data. This modeling can be categorized into heuristic and metaheuristic algorithms accord-
technique is common across the literature and is applied to predict
ing to their core learning strategy. Heuristics are problem-dependent
single-blast or multiple-blast objectives.
strategies, meaning they are tailored to the specific problem at hand.
Optimization. In addition to prediction, some researchers have ex- Metaheuristics are problem-independent strategies that can be adapted
tended their ML models to optimize the input parameters for improved to a range of different problems (Abdel-Basset, Abdel-Fatah, & Sanga-
blast designs. This process is done in two stages. Firstly, a base model iah, 2018; Desale, Rasool, Andhale, & Rane, 2015), such as evolutionary
is created and then used to develop a second model via a hybrid algorithms.
approach. For example, Bayat et al. (2022) developed a gene expression
Hybrid approach. The hybrid approach aims to enhance the accuracy
programming model to predict ground vibration with a high level of
of predictions or optimizations by combining two ML models, which
accuracy and used this as an objective function for the grasshopper
could either be similar or different (Armaghani et al., 2020; Fang
optimization algorithm to optimize the input blast parameters.
et al., 2021). By leveraging the strengths and compensating for the
6.2.2. Model development approach weaknesses of each model, a hybrid approach leads to improved overall
The application of ML models in drilling and blasting has seen a model performance. For instance, while one model might be proficient
balanced exploration between individual and hybrid approaches, with at capturing linear relationships, another might excel at detecting non-
50% (104) of the studies adopting each method. Our analysis of the linear relationships. By combining these models, the resulting hybrid
literature reveals that in the early years, there was a greater focus model can capture both types of relationships in the data to enhance
on individual models with less exploration of hybrid models. This prediction accuracy (Hasanipanah et al., 2018). This hybrid approach
trend is illustrated in Fig. 7, where the bar chart shows the year-wise can be achieved by combining: (1) two heuristic models (Nguyen & Bui,
distribution of hybrid and single models. The focus on hybrid models 2019), (2) two metaheuristic models (Zhou, Koopialipoor, et al., 2020), or
began in 2011, with a pronounced shift occurring in 2020, surpassing (3) a heuristic and a metaheuristic model (Sayevand et al., 2018). In each
individual models in the percentage of studies conducted. This suggests case, hybridization is realized either by optimizing the hyperparameters
a growing preference for hybrid approaches. of one model (referred to as the base model) using another ML model
The line chart in Fig. 8 further reinforces this analysis by showing or by developing a base model and using it as an objective function of
the cumulative percentage of studies dedicated to both hybrid and another ML model.
single models across the years. The year 2018 marks a turning point For example, Bui et al. (2021) achieved hybridization by opti-
for individual model studies, with 50% of the studies conducted before mizing the weights and biases of artificial neural networks using a
this year. Conversely, the year 2020 stands out as a milestone for the cuckoo search optimizer. Whereas, Rezaeineshat et al. (2020) achieved
hybrid model studies by which half of the studies had been conducted hybridization by developing an artificial neural network model for
by this point. The sharp rise in the cumulative percentage of hybrid predicting blast-induced ground vibration and then using this model
model studies since 2020 further emphasizes the increasing interest in as an objective function for an imperialist competitive algorithm to
multifaceted ML models during recent years. optimize blast parameters.

11
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Fig. 8. Cumulative percentage of studies for hybrid and single models.

Fig. 9. Word cloud of ML models across literature.

The hybrid approach provides a strategy for improving prediction (78% studies) that uses a single metric to compare the model’s per-
accuracy; however, it is not the only method that combines multiple ML formance against a different model on the same dataset (Bui et al.,
models. An ensemble approach involves combining multiple models to 2021; Gorai et al., 2021; Monjezi et al., 2021; Xie et al., 2021; Zeng
achieve higher accuracy compared to their individual performances (He et al., 2022). The least common is the ranking system (22% studies)
et al., 2021). Despite both methods combining different ML models, hy- that assigns relative ranks to each model based on its performance
brid and ensemble approaches differ in the way they handle the outputs across multiple metrics (Barkhordari et al., 2022; Chen, Armaghani,
of their individual models. Firstly, an ensemble approach aggregates et al., 2022; Yu, Shi, Zhou, Gou, et al., 2020; Yu, Shi, Zhou, et al.,
the outputs of the individual models known as base learners, while a 2021). This approach recognizes that a single metric may not capture
hybrid approach uses the output from a single model as input to the all the essential characteristics of a model’s performance, enabling a
other model. Secondly, the base learners in ensembles are trained on comprehensive evaluation strategy.
unique data subsets from the original dataset, while hybrid models use
a single dataset to train their individual models. 7. Discussion
The word cloud in Fig. 9 presents the central theme of surveyed
ML models, emphasizing their versatility and widespread application in Research into optimizing drill and blast design parameters is cat-
drilling and blasting. It illustrates a preference for neural networks, sup- egorized based on their objectives to maximize or minimize specific
port vector machines, and genetic algorithms, due to their robustness outcomes. We identified three main categories of studies during our lit-
in handling optimization problems. Furthermore, to tackle the dynamic erature review with the objectives of: (i) production, (ii) environmental
challenges inherent in the field of drilling and blasting, metaheuristic safety, and (iii) rock stability. The first category focuses on maximizing
algorithms were adapted for hybridization model development, such as production from the drilling and blasting phase to generate revenue
particle swarm, firefly and whale optimization algorithms. for mines. Rock fragmentation size and throw distance are important
parameters in determining the operational costs of mining and are di-
6.3. Evaluation rectly associated with mineral production and reducing ore loss during
blasting. The second category aims to minimize adverse environmental
A range of evaluation metrics is used based on the task the model impacts such as ground vibration, flyrock, and air overpressure. These
performs (the model type), for example, if it is a regression or classifica- outcomes are considered undesirable as they have a negative impact on
tion model. We identified 40 distinct evaluation metrics for measuring the environment and human safety. These outcomes have a threshold
the regression model’s performance. The most common were coefficient limit that is dependent on the site conditions and is set based on factors
of determination (𝑅2 ), root mean square error, variance account for, beyond the mine engineer’s control. The final category includes studies
and mean absolute error. The two studies that used classification mod- related to rock stability, such as backbreak and toe formation. These
els used accuracy to evaluate model performance (Mohammadi et al., are also considered undesirable as they can impact the operational
2020; Zhou, Asteris, et al., 2020). After evaluating model performance, procedure/s and costs of the subsequent blasting operation.
researchers used one of two main methods to compare the performance The three objectives are site-specific. In cases where mines are close
of different models. The most common is the single metric evaluation to a habitat, there needs to be a balance between the three objectives.

12
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Table 4
Issues found in various stages of ML model development and our recommendations for future research.
Stage Issues Recommendation
Data preprocessing - Availability of limited real-world data - Developing a data schema for consistent and
- Data considered for model development is not accurate data collection and curation
representative of the actual drill and blast design - An approach to collect and integrate
- Feature selection is at author’s discretion uncontrollable factors (geological parameters) with
- Exclusion of geological parameters drill and blast design data
- Ambiguity in the choice of preprocessing methods - A data pipeline for streamlining and automating
data preprocessing
Model selection & training - Model selection ambiguity as many models have - Focus on multiobjective models with an aim to
the same level of prediction accuracy optimize the input feature values
- Lack of adequate focus on multiobjective studies - ML pipeline for systematically comparing and
- Focus is on prediction, not input optimization selecting the best ML model that can satisfy the
constraints in the case of multiobjective models
Evaluation - Lack of quantitative method to evaluate the - Developing a weighted mathematical model for
performance against all three objectives comprehensive evaluation of drilling and blasting
- Absence of testing in the real-world setting outcomes

However, sites that are isolated can work on trade-offs between the but different dataset can overcome the problem of small datasets to
three objectives. Traditional methods rely on manual effort; thus, it improve model performance.
is not feasible to find a balance between the three objectives, and it In addition, data quality in research is critical. The data should be
can be expensive and time-consuming to achieve multiple objectives reliable, accurate, and valid. However, despite researchers providing
from a single drill and blast. As a result, no common method exists that information on the source of the data, there is a lack of information
can be adapted across mines to achieve these objectives. Our analysis on how the data was collected. Thus, the data accuracy and ML model
of the current studies revealed that most authors focus on developing results are questionable. Drilling is a dynamic process where deviations
ML models for drill and blast design optimization. However, we found from the initial drilling plan can occur due to geological inconsistencies
that many of these studies focus on a single objective, and only a small or human error. These deviations can impact the performance of the
number (6% of studies) have attempted to address multiple objectives. blast design. However, none of the literature has provided information
In attaining these objectives, we identified several issues in the ML on any deviations or mentioned whether the data was collected before
model development process from the authors. While some of these or after drilling. This lack of information challenges the validity of
issues require changes in mine operating procedures, such as data the data and highlights the need for more robust data collection and
collection methods, others can be addressed through changes in the validation methods in future studies. Documentation of data collection
ML model development workflow. Therefore, we conducted a compre- methods and any deviations from the plan or implementation of a data
hensive analysis of the current research to identify the challenges and schema or a pipeline can ensure the validity and reliability of the data.
issues in developing ML models for drill and blast design optimization,
which we have presented in this section. Table 4 presents a summarized 7.2. Data preparation
view of the various stages in ML model development, the associated
issues, and our recommendations for future research. This overview The inconsistency and ambiguity in data preprocessing across lit-
allows researchers to recognize and understand the issues in the current erature can lead to biased or inaccurate model parameters and poor
literature and offers guidance for developing more robust and reliable performance. We found that it is common practice for researchers not to
ML models for drilling and blasting applications. clarify the cleaning, transformation, and reduction techniques used and
found the selection criteria ambiguous. For example, Jahed Armaghani
7.1. Datasets et al. (2018) and Jahed Armaghani, Hajihassani, Marto, et al. (2015)
follow different approaches in handling outliers, despite being by the
We have identified data collection as a major challenge in the field same author. Zhou, Qiu, et al. (2021) has chosen multiple input factors
of drilling and blasting. This challenge arises from the complexities of but failed to perform dimensionality reduction, resulting in the pres-
gathering sufficient data, such as safety considerations, accessibility of ence of highly correlated factors such as hole depth and charge length.
relevant sites, and the specialized nature of required information. 65% This can lead to overfitting and exaggerating the model’s performance.
(135 out of 208) studies have used datasets with 150 or fewer records To address these issues, a standardized data pipeline is needed to
to train the ML models. This small sample size can impact the model’s streamline and automate data collection and preprocessing steps and
performance leading to overfitting or underfitting. This small dataset document the methods used in each step, ensuring that the data used
problem can be alleviated (to an extent) by using techniques like in future studies are reliable, valid, and representative of the problem.
data augmentation, synthetic data generation, and transfer learning. Another challenge is data representation. We found that none of the
For example, Mohammadi et al. (2020), Monjezi, Mehrdanesh, et al. existing literature reported the blast record as an actual representation
(2013), Ghasemi (2017), and Mohammadnejad et al. (2013) have used of the drill and blast design. Researchers aggregated values of the
datasets collected from the Sungun copper mine; however, they differ in design parameters as input for ML model training. However, using
size and features. It is worth noting that while Ghasemi (2017) and Mo- aggregate values of the data does not provide a complete and accurate
hammadnejad et al. (2013) aims to predict backbreak, Mohammadi picture of the design. This aggregation may predict the objective; how-
et al. (2020) and Monjezi, Mehrdanesh, et al. (2013) focus on predicting ever, it does not assist in blast parameter optimization. A typical blast
ground vibration and flyrock, respectively. Despite the difference in involves multiple rows where each row has multiple holes. Parameters
prediction variables, both studies focus on optimizing the drill and such as spacing, burden, hole depth, explosive amount, and stemming
blast design. In this case, transfer learning can be an effective approach differ from hole to hole and row to row, resulting in multiple values
for faster convergence and robust model development. Leveraging the per parameter in a single blast design. We found that only 4% (8 out
knowledge of a pre-trained model on one dataset and using it to of 208) of studies reported taking a mean or average value for these
establish initial parameters for training the same model on a similar parameters. The remaining 200 studies have not reported how they

13
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

handled this data, raising questions about the trust and accuracy of 7.5. Performance evaluation & real world application
their study results. To account for this variability, we recommend using
a database schema that captures the important data of a blast design. In the mining industry, mine engineers evaluate drill and blast
design as a measure of the achievability of mine-specific goals such as
7.3. Data leakage production or cost or environmental compliance, or a combination of
all three. Despite the critical importance of these objectives, a quan-
Data leakage is a common problem in developing ML models. It titative method for comprehensive evaluation of model performance
occurs when unwanted data is used in training the model. Kapoor and against all three objectives is lacking. Standard statistical metrics are
Narayanan (2022) has defined data leakage as a spurious relationship commonly used to measure model performance based on a particular
between the independent variables and the target variable that arises test dataset. Moreover, the high cost of drilling and blasting limits real-
as an artifact of the data collection, sampling, or preprocessing strategy world experiments. As previously noted, most studies have primarily
leading to inflated model performance estimates. Data leakage is intro- focused on building predictive models. Although these models can
duced in the ML model development lifecycle during the preprocessing assist in predicting the outcomes for a specific drill and blast design,
stage. A common cause for data leakage is performing imputation they rely on the mine engineer to input design values. To reduce this
or normalization before the data split, allowing the model to learn reliance and to improve the practical application of ML models, there
not only from the training dataset but also from the test dataset. is a need for more research to focus on multiobjective optimization
This preprocessing leakage observed in several studies has resulted in methods. This approach would provide mine engineers with the optimal
over-optimism of the model prediction performance (Ke et al., 2021; values for drill and blast design parameters, reducing the reliance on
Nguyen, Bui, Nguyen-Thoi, et al., 2019; Ramesh Murlidhar et al., 2021; human decision-making and paving the way for implementing auto-
Zhang, Bui, et al., 2020). To minimize the data leakage issue, one mated technologies in the future. Although current evaluation methods
approach is to split the data into train, validate, and test datasets and can demonstrate the potential of ML for real-world applications, there
use cross-validation in the training process when the amount of data is room for improvement by focusing on multiobjective optimization.
available is limited. Another approach is to hold back a separate dataset Multiobjective optimization can provide more comprehensive evalua-
for validation to help perform a sanity check to identify the case of tion and optimization solutions, making ML models even more useful
overfitting. Insufficient information in study reports, coupled with the in practical settings.
issue of data leakage, can lead to concerns about reproducibility, where
the reliability and validity of the results are questionable. 8. Conclusion

7.4. Underspecification Advances in technology have led to much attention in ML model


development in drilling and blasting. This review shows promising
The transformation of these studies into real-world applications outcomes from research in ML models to optimize the drill and blast
is limited due to underspecification. This term, as defined by Teney, design. We begin with an introduction to the traditional methods of
Peyrard, and Abbasnejad (2022), refers to the gap between the in- developing drill and blast designs for rock blasting and its associated
formation provided in a dataset used for model development and the complexities, followed by a detailed approach used by researchers to
information required to perform effectively on a task in a real-world set- develop and apply ML models in this context.
ting. Specifically, underspecification arises when the dataset used is not Drill and blast designs are influenced by controllable and uncon-
representative of the target environment, making it challenging for the trollable factors. Data collection methods vary across these factors
model to learn the underlying patterns in the data and generalize well and are currently manually collected, resulting in limited data avail-
to unseen data, resulting in poor out-of-sample performance (D’Amour ability. Studies identified in this survey used datasets with 100 or
et al., 2022). In our analysis, we found that 63% (130 out of 208) fewer records. This highlights the challenges and opportunities for
of the studies overlooked the variability of mine conditions, result- ML model development. Traditionally, mine engineers relied on the
ing in a mismatch between the study and deployment environments. empirical evidence collected at a particular mine to develop drill and
Consequently, despite the development of accurate models in these blast designs. These methods are prone to error, unreliable, time-
studies, their application might not be reliable or easily generalizable to intensive, and expensive. As a result, researchers are leveraging ML
different mine sites with varying geological conditions. This selection models to find reliable and faster methods for developing optimal
bias is caused by relying on individual experience when choosing input drill and blast designs. However, a critical step in this process is data
features and further affects the model’s credibility. preprocessing, including data cleaning and dimensionality reduction.
Addressing underspecification requires reducing the discrepancy Our survey revealed that only 57% of the studies adequately addressed
between the training environment and the deployment environment. data preparation, with a notable lack of attention to outlier removal
One approach uses a dataset for training and testing that accurately and dimensionality reduction, which are crucial for the accuracy and
reflects the potential heterogeneity of the mining conditions, leading to reliability of ML models.
models that are capable of handling the diverse scenarios encountered In this paper, we analyzed research conducted on ML models for
in the real world. Another approach is to introduce inductive biases drill and blast design optimization between 2005 and 2022. To the best
into the training process (Mitchell, 1980a; Wolpert, 1996). Inductive of our knowledge, this is the first survey to provide a comprehensive
biases can come in various forms, including incorporating domain review and critical analysis of the ML model development process in
knowledge into the model, taking into account site-specific constraints, drilling and blasting. Our survey revealed that the primary focus of
or using regularization techniques. This necessitates close collaboration the ML models was to predict drilling and blasting outcomes, such as
between data scientists and mine engineers to ensure that the models rock fragmentation, flyrock, ground vibration, or air overpressure. The
are grounded in real-world knowledge and constraints. The inclusion of development of these models has shifted from individual approaches
domain knowledge and expert input can lead to more realistic model- to hybrid methods, aiming to predict outcomes from historical data
ing scenarios and improved generalizability of the models (Holzinger, and optimize input parameters. The effectiveness of these models was
Biemann, Pattichis, & Kell, 2017; Mitchell, 1980b). By employing these assessed based on the accuracy of their predictions for a specific drill
strategies, it is possible to bridge the gap between the information used and blast design. This evaluation employed various metrics, the most
for training the model and the information required to make accurate common being coefficient of determination (𝑅2 ), root mean square
predictions or optimizations in a real-world setting. error, and mean absolute error. Two main methods to compare model

14
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Table A.1
List of abbreviations.
Abbreviation Full Form Abbreviation Full Form
COV Coefficient of Variation FA Firefly Algorithm
NSE Nash and Sutcliffe statistic PSO Particle Swarm Optimizer
MABE Mean Absolute Bias Error WO Whale Optimizer
MEDAE Median Absolute Error ICA Imperialist Competitive Algorithm
𝑅2 Coefficient of Determination JSO Jellyfish Search Optimizer
RMSE Root Mean Square Error HHO Harris Hawks Optimizer
MAE Mean Absolute Error MFO Moth Flame Optimizer
MSE Mean Square Error ABC Artificial Bee Colony Optimizer
SS Skill Score EO Equilibrium Optimizer
PRMSE Predictive Root Mean Square Error BO Biogeography-based Optimizer
MAPE Mean Absolute Percentage Error HLO Human Learning Optimizer
SEP Standard Error of Prediction JA Jaya Algorithm
Er Mean Relative Error SSO Sparrow Search Optimizer
NRMSE Normalized Root Mean Square Error MRFO Manta Ray Foraging Optimizer
AARE Average Absolute Relative Error SFS Stochastic Fractal Search
RSE Root Squared Error ANN Artificial Neural Network
U1 Prediction Accuracy BIENN Brain-Inspired Emotional Neural Network
U2 Prediction Quality CFNN Cascaded Forward Neural Network
Cp Coefficient of Persistence CWNN Casualty Weighted Neural Network
d Modified Agreement Index DNN Deep Neural Network
SMAPE Systematic Mean Absolute Percentage Error RFNN Recurrent Fuzzy Neural Network
LM Leegate and McCabe Index BNN Bayesian Network Technique
WI Willmott’s Index of Agreement AENN Autoencoder Neural Network
RMSER Hybrid Performance Index QRNN Quantile Regression Neural Network
MBE Mean Average Error SONN Self-Organizing Neural Network
VARE Variance Absolute Relative Error GP Genetic Programming
SSE Sum of Squares Error GMDH Group Method of Data Handling
MMRE Mean Magnitude of Relative Error CA Cubist Algorithm
FB Fractional Bias BSS Boosted Smoothing Spline
PI Performance Index RF Random Forest
MRE Maximum Relative Error GA Genetic Algorithm
NMSE Normalized Mean Square Error GEP Gene Expression Programming
SI Scatter Index RBFN Radial Bias Function Network
CRM Coefficient of Residual Index SVM Support Vector Machine
U95 Uncertainty at 95% ELM Extreme Learning Machine
BIC Bayesian Information Criterion MCS Monte Carlo Simulation
Std_AE Standardized Absolute Error ADHS Adaptive Dynamical Harmony Search
Std_APE Standardized Absolute Percentage Error RVM Relevance Vector Machine
GBM Gradient Boosting Machine HPSOGWO Hybrid Particle Swarm Optimization with Grey Wolf Optimizer
BGAM Boosted Generalized Additive Model CRANFIS Chaos Recurrent Adaptive Neuro-Fuzzy Inference System
GWO Grey Wolf Optimizer IM Itemset Mining
GO Grasshopper Optimizer HKM Hierarchical k-Means Clustering
SVM Support Vector Machine FCM Fuzzy C-Means Clustering
CSO Cat Swarm Optimizer KNN k-Nearest Neighbors
CO Cuckoo Search Optimizer CART Classification and Regression Tree
ACO Ant Colony Optimizer MARS Multivariate Adaptive Regression Splines
RT Regression Tree SCA Sine Cosine Algorithm
BRT Boosted Regression Tree RSM Response Surface Model
GPR Gaussian Process Regression
FIS Fuzzy Inference System
ANFIS Adaptive Neural Fuzzy Inference System

performance are single metric evaluation and the ranking system, directly influence drill and blast parameters rather than mere predictive
allowing for a more comprehensive evaluation. models.
In conclusion, our survey on ML models for drill and blast design
optimization shows that while the results of these studies are promising, CRediT authorship contribution statement
developing such models is still challenging due to the ambiguity in
ML methods and dynamic mine conditions. Key challenges include Venkat Munagala: Conceptualization, Methodology, Investigation,
data leakage, underspecification, and lack of comprehensive evaluation Formal analysis, Data curation, Writing – original draft, Visualiza-
methods. More research is needed to develop a structured and reusable tion. Srikanth Thudumu: Conceptualization, Methodology, Writing
approach for comparing different models and selecting the best one for – review & editing, Supervision. Irini Logothetis: Conceptualization,
specific site constraints. We recommend developing an ML pipeline to Methodology, Writing – review & editing, Supervision. Sushil Bhan-
facilitate this comparison and improve the reliability and efficiency of dari: Conceptualization, Methodology, Validation. Rajesh Vasa: Con-
drill and blast design optimization. Future research directions should ceptualization, Methodology, Writing – review & editing, Supervision.
focus on bridging the gap between academic research and indus- Kon Mouzakis: Conceptualization, Methodology, Writing – review &
trial application, emphasizing the development of methodologies that editing, Supervision.

15
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Declaration of competing interest Armaghani, D. J., Momeni, E., Abad, S. V. A. N. K., & Khandelwal, M. (2015). Feasibility
of ANFIS model for prediction of ground vibrations resulting from quarry blasting.
Environmental Earth Sciences, 74(4), 2845–2860.
The authors declare that they have no known competing finan-
Arthur, C. K., Bhatawdekar, R. M., Mohamad, E. T., Sabri, M. M. S., Bohra, M.,
cial interests or personal relationships that could have appeared to Khandelwal, M., et al. (2022). Prediction of blast-induced ground vibration at a
influence the work reported in this paper. limestone quarry: An artificial intelligence approach. Applied Sciences, 12(18), 9189.
Asl, P. F., Monjezi, M., Hamidi, J. K., & Armaghani, D. J. (2018). Optimization of
flyrock and rock fragmentation in the Tajareh limestone mine using metaheuristics
Data availability method of firefly algorithm. Engineering with Computers, 34(2), 241–251.
Asr, E. T., Kakaie, R., Ataei, M., & Mohammadi, M. R. T. (2019). A review of studies
The data used in this article is publicly available. on sustainable development in mining life cycle. Journal of Cleaner Production, 229,
213–231.
Azimi, Y., Khoshrou, S. H., & Osanloo, M. (2019). Prediction of blast induced ground
Acknowledgments vibration (BIGV) of quarry mining using hybrid genetic algorithm optimized
artificial neural network. Measurement, 147, Article 106874.
This work is supported as part of the Higher Degree Research (HDR) Babaeian, M., Sereshki, F., Ataei, M., Nehring, M., & Mohammadi, S. (2023). Appli-
cation of soft computing, statistical and multi-criteria decision-making methods to
program at the Applied Artificial Intelligence Institute (𝐴2 𝐼 2 ), Deakin
develop a predictive equation for prediction of flyrock distance in open-pit mining.
University. We thank MineExcellence for their valuable Subject Matter Mining, 3(2), 304–333.
Expertise (SME) support. Bahrami, A., Monjezi, M., Goshtasbi, K., & Ghazvinian, A. (2011). Prediction of rock
fragmentation due to blasting using artificial neural network. Engineering with
Computers, 27(2), 177–181.
Appendix. Abbreviations
Bakhtavar, E., Sadiq, R., & Hewage, K. (2021). Optimization of blasting-associated costs
in surface mines using risk-based probabilistic integer programming and firefly
See Table A.1. algorithm. Natural Resources Research, 30(6), 4789–4806.
Bamford, T., Esmaeili, K., & Schoellig, A. P. (2021). A deep learning approach for rock
fragmentation analysis. International Journal of Rock Mechanics and Mining Sciences,
145, Article 104839.
References Barkhordari, M., Armaghani, D., & Fakharian, P. (2022). Ensemble machine learning
models for prediction of flyrock due to quarry blasting. International Journal of
Abbaspour, H., Drebenstedt, C., Badroddin, M., & Maghaminik, A. (2018). Optimized Environmental Science and Technology, 1–16.
design of drilling and blasting operations in open pit mines under technical and Bayat, P., Monjezi, M., Mehrdanesh, A., & Khandelwal, M. (2022). Blasting pattern
economic uncertainties by system dynamic modelling. International Journal of Mining optimization using gene expression programming and grasshopper optimization
Science and Technology, 28(6), 839–848. algorithm to minimise blast-induced ground vibrations. Engineering with Computers,
Abd Elwahab, A., Topal, E., & Jang, H. D. (2023). Review of machine learning 38(4), 3341–3350.
application in mine blasting. Arabian Journal of Geosciences, 16(2), 133. Bayat, P., Monjezi, M., Rezakhah, M., & Armaghani, D. J. (2020). Artificial neural
Abdel-Basset, M., Abdel-Fatah, L., & Sangaiah, A. K. (2018). Metaheuristic algorithms: network and firefly algorithm for estimation and minimization of ground vibration
A comprehensive review. Computational Intelligence for Multimedia Big Data on the induced by blasting in a mine. Natural Resources Research, 29(6), 4121–4132.
Cloud with Engineering Applications, 185–231. Behzadafshar, K., Mohebbi, F., Tehrani, M. S., Hasanipanah, M., & Tabrizi, O.
Aggarwal, C. C., & Yu, P. S. (2001). Outlier detection for high dimensional data. In (2018). Predicting the ground vibration induced by mine blasting using imperialist
Proceedings of the 2001 ACM SIGMOD international conference on management of data competitive algorithm. Engineering Computations.
(pp. 37–46). Bhandari, S. (1997). Engineering rock blasting operations. Rotterdam: AA Balkema.
Al-Bakri, A. Y., & Sazid, M. (2021). Application of artificial neural network (ANN) for Bhatawdekar, R. M., Kumar, R., Sabri Sabri, M. M., Roy, B., Mohamad, E. T., Kumar, D.,
prediction and optimization of blast-induced impacts. Mining, 1(3), 315–334. et al. (2023). Estimating flyrock distance induced due to mine blasting by extreme
learning machine coupled with an equilibrium optimizer. Sustainability, 15(4), 3265.
Ali, D., & Frimpong, S. (2020). Artificial intelligence, machine learning and process
Bilim, N., Çelik, A., & Kekeç, B. (2017). A study in cost analysis of aggregate production
automation: Existing knowledge frontier and way forward for mining sector.
as depending on drilling and blasting design. Journal of African Earth Sciences, 134,
Artificial Intelligence Review, 53(8), 6025–6042.
564–572.
Amershi, S., Begel, A., Bird, C., DeLine, R., Gall, H., Kamar, E., et al. (2019). Software
Bołoz, Ł., & Biały, W. (2020). Automation and robotization of underground mining in
engineering for machine learning: A case study. In 2019 IEEE/ACM 41st international
Poland. Applied Sciences, 10(20), 7221.
conference on software engineering: Software engineering in practice (pp. 291–300).
Bui, X.-N., Choi, Y., Atrushkevich, V., Nguyen, H., Tran, Q.-H., Long, N. Q., et al.
IEEE.
(2020). Prediction of blast-induced ground vibration intensity in open-pit mines
Amini, H., Gholami, R., Monjezi, M., Torabi, S. R., & Zadhesh, J. (2012). Evaluation of
using unmanned aerial vehicle and a novel intelligence system. Natural Resources
flyrock phenomenon due to blasting operation by support vector machine. Neural
Research, 29(2), 771–790.
Computing and Applications, 21(8), 2077–2085.
Bui, X. N., Jaroonpattanapong, P., Nguyen, H., Tran, Q. H., & Long, N. Q. (2019).
AminShokravi, A., Eskandar, H., Derakhsh, A. M., Rad, H. N., & Ghanadi, A. (2018).
A novel hybrid model for predicting blast-induced ground vibration based on
The potential application of particle swarm optimization algorithm for forecasting
k-nearest neighbors and particle swarm optimization. Scientific Reports, 9(1), 1–14.
the air-overpressure induced by mine blasting. Engineering with Computers, 34(2),
Bui, X. N., Nguyen, H., Le, H. A., Bui, H. B., & Do, N. H. (2020). Prediction of
277–285.
blast-induced air over-pressure in open-pit mine: Assessment of different artificial
Amiri, M., Bakhshandeh Amnieh, H., Hasanipanah, M., & Mohammad Khanli, L. (2016). intelligence techniques. Natural Resources Research, 29(2), 571–591.
A new combination of artificial neural network and K-nearest neighbors models Bui, X. N., Nguyen, H., Tran, Q. H., Nguyen, D. A., & Bui, H.-B. (2021). Predicting
to predict blast-induced ground vibration and air-overpressure. Engineering with ground vibrations due to mine blasting using a novel artificial neural network-based
Computers, 32(4), 631–644. cuckoo search optimization. Natural Resources Research, 30(3), 2663–2685.
Amiri, M., Hasanipanah, M., & Bakhshandeh Amnieh, H. (2020). Predicting ground Chandrahas, N. S., Choudhary, B. S., Teja, M. V., Venkataramayya, M., & Prasad, N. K.
vibration induced by rock blasting using a novel hybrid of neural network and (2022). XG boost algorithm to simultaneous prediction of rock fragmentation and
itemset mining. Neural Computing and Applications, 32(18), 14681–14699. induced ground vibration using unique blast data. Applied Sciences, 12(10), 5269.
Amoako, R., Jha, A., & Zhong, S. (2022). Rock fragmentation prediction using an Chen, L., Armaghani, D., Fakharuab, P., Bhatawdekar, R., Samui, P., Khandelwal, M.,
artificial neural network and support vector regression hybrid approach. Mining, et al. (2022). A study on environmental issues of blasting using advanced support
2(2), 233–247. vector machine algorithms. International Journal of Environmental Science and
Arlot, S., & Celisse, A. (2010). A survey of cross-validation procedures for model Technology, 1–20.
selection. Statistics Surveys, 4(none), 40–79. http://dx.doi.org/10.1214/09-SS054. Chen, L., Asteris, P. G., Tsoukalas, M. Z., Armaghani, D. J., Ulrikh, D. V., & Yari, M.
Armaghani, D. J., Hajihassani, M., Mohamad, E. T., Marto, A., & Noorani, S. (2014). (2022). Forecast of airblast vibrations induced by blasting using support vector
Blasting-induced flyrock and ground vibration prediction through an expert arti- regression optimized by the grasshopper optimization (SVR-GO) technique. Applied
ficial neural network based on particle swarm optimization. Arabian Journal of Sciences, 12(19), 9805.
Geosciences, 7(12), 5383–5396. Chen, W., Hasanipanah, M., Nikafshan Rad, H., Jahed Armaghani, D., & Tahir, M.
Armaghani, D. J., Hasanipanah, M., Amnieh, H. B., & Mohamad, E. T. (2018). (2021). A new design of evolutionary hybrid optimization of SVR model in
Feasibility of ICA in approximating ground vibration resulting from mine blasting. predicting the blast-induced ground vibration. Engineering with Computers, 37(2),
Neural Computing and Applications, 29(9), 457–465. 1455–1471.
Armaghani, D. J., Koopialipoor, M., Bahri, M., Hasanipanah, M., & Tahir, M. (2020). A Chu, X., Ilyas, I. F., Krishnan, S., & Wang, J. (2016). Data cleaning: Overview
SVR-GWO technique to minimize flyrock distance resulting from blasting. Bulletin and emerging challenges. In Proceedings of the 2016 international conference on
of Engineering Geology and the Environment, 79(8), 4369–4385. management of data (pp. 2201–2206).

16
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Cunningham, C. (1983). The Kuz-Ram model for prediction of fragmentation from Gao, W., Alqahtani, A. S., Mubarakali, A., Mavaluru, D., et al. (2020). Developing an
blasting. In Proc. first int. symp. on rock fragmentation by blasting (pp. 439–453). innovative soft computing scheme for prediction of air overpressure resulting from
Dai, Y., Khandelwal, M., Qiu, Y., Zhou, J., Monjezi, M., & Yang, P. (2022). A mine blasting using GMDH optimized by GA. Engineering with Computers, 36(2),
hybrid metaheuristic approach using random forest and particle swarm optimization 647–654.
to study and evaluate backbreak in open-pit blasting. Neural Computing and Gao, W., Karbasi, M., Hasanipanah, M., Zhang, X., & Guo, J. (2018). Developing GPR
Applications, 34(8), 6273–6288. model for forecasting the rock fragmentation in surface mines. Engineering with
D’Amour, A., Heller, K., Moldovan, D., Adlam, B., Alipanahi, B., Beutel, A., et al. (2022). Computers, 34(2), 339–345.
Underspecification presents challenges for credibility in modern machine learning. García, S., Luengo, J., & Herrera, F. (2015). Data preprocessing in data mining, vol. 72.
Journal of Machine Learning Research, 23(1), 10237–10297. Springer.
de Souza Nascimento, E., Ahmed, I., Oliveira, E., Palheta, M. P., Steinmacher, I., & Ghahramani, Z. (2003). Unsupervised learning. In Summer school on machine learning
Conte, T. (2019). Understanding development process of machine learning systems: (pp. 72–112). Springer.
Challenges and solutions. In 2019 ACM/IEEE international symposium on empirical Ghasemi, E. (2017). Particle swarm optimization approach for forecasting backbreak
software engineering and measurement (pp. 1–6). IEEE. induced by bench blasting. Neural Computing and Applications, 28(7), 1855–1862.
Dehghani, H., Pourzafar, M., et al. (2021). Prediction and minimization of blast-induced Ghasemi, E., Amini, H., Ataei, M., & Khalokakaei, R. (2014). Application of artificial
flyrock using gene expression programming and cuckoo optimization algorithm. intelligence techniques for predicting the flyrock distance caused by blasting
Environmental Earth Sciences, 80(1), 1–17. operation. Arabian Journal of Geosciences, 7(1), 193–202.
Desale, S., Rasool, A., Andhale, S., & Rane, P. (2015). Heuristic and meta-heuristic Ghasemi, E., Amnieh, H. B., & Bagherpour, R. (2016). Assessment of backbreak due
algorithms and their relevance to the real world: A survey. International Journal of to blasting operation in open pit mines: A case study. Environmental Earth Sciences,
Computer Engineering in Research Trends, 351(5), 2349–7084. 75(7), 1–11.
Dimitraki, L., Christaras, B., Marinos, V., Vlahavas, I., & Arampelos, N. (2019). Predict- Ghasemi, E., Ataei, M., & Hashemolhosseini, H. (2013). Development of a fuzzy model
ing the average size of blasted rocks in aggregate quarries using artificial neural for predicting ground vibration caused by rock blasting in surface mining. Journal
networks. Bulletin of Engineering Geology and the Environment, 78(4), 2717–2729. of Vibration and Control, 19(5), 755–770.
Ding, X., Hasanipanah, M., Nikafshan Rad, H., & Zhou, W. (2021). Predicting the blast- Gorai, A. K., Himanshu, V. K., & Santi, C. (2021). Development of ANN-based universal
induced vibration velocity using a bagged support vector regression optimized with predictor for prediction of blast-induced vibration indicators and its performance
firefly algorithm. Engineering with Computers, 37(3), 2273–2284. comparison with existing empirical models. Mining, Metallurgy & Exploration, 38(5).
Ding, X., Jamei, M., Hasanipanah, M., Abdullah, R. A., & Le, B. N. (2023). Optimized Görgülü, K., Arpaz, E., Demirci, A., Koçaslan, A., Dilmaç, M. K., & Yüksek, A. G. (2013).
data-driven models for prediction of flyrock due to blasting in surface mines. Investigation of blast-induced ground vibrations in the Tülü boron open pit mine.
Sustainability, 15(10), 8424. Bulletin of Engineering Geology and the Environment, 72(3), 555–564.
Ding, Z., Nguyen, H., Bui, X. N., Zhou, J., & Moayedi, H. (2020). Computational Guo, H., Nguyen, H., Bui, X. N., & Armaghani, D. J. (2021). A new technique to predict
intelligence model for estimating intensity of blast-induced ground vibration in a fly-rock in bench blasting based on an ensemble of support vector regression and
mine based on imperialist competitive and extreme gradient boosting algorithms. GLMNET. Engineering with Computers, 37(1), 421–435.
Natural Resources Research, 29(2), 751–769.
Guo, Q., Yang, S., Wang, Y., & Xiang, Z. (2022). Prediction research for blasting
Dumakor-Dupey, N. K., Arya, S., & Jha, A. (2021). Advances in blast-induced impact
peak particle velocity based on random GA-BP network group. Arabian Journal
prediction—A review of machine learning applications. Minerals, 11(6), 601.
of Geosciences, 15(15), 1351.
Ebrahimi, E., Monjezi, M., Khalesi, M. R., & Armaghani, D. J. (2016). Prediction
Guo, H., Zhou, J., Koopialipoor, M., Jahed Armaghani, D., & Tahir, M. (2021). Deep
and optimization of back-break and rock fragmentation using an artificial neural
neural network and whale optimization algorithm to assess flyrock induced by
network and a bee colony algorithm. Bulletin of Engineering Geology and the
blasting. Engineering with Computers, 37(1), 173–186.
Environment, 75(1), 27–36.
Hajihassani, M., Armaghani, D. J., Sohaei, H., Mohamad, E. T., & Marto, A. (2014).
Enayatollahi, I., Aghajani Bazzazi, A., & Asadi, A. (2014). Comparison between neural
Prediction of airblast-overpressure induced by blasting using a hybrid artificial
networks and multiple regression analysis to predict rock fragmentation in open-pit
neural network and particle swarm optimization. Applied Acoustics, 80, 57–67.
mines. Rock Mechanics and Rock Engineering, 47(2), 799–807.
Hajihassani, M., Jahed Armaghani, D., Marto, A., & Tonnizam Mohamad, E. (2015).
Eskandar, H., Heydari, E., Hasanipanah, M., Masir, M. J., & Derakhsh, A. M. (2018).
Ground vibration prediction in quarry blasting through an artificial neural network
Feasibility of particle swarm optimization and multiple regression for the prediction
optimized by imperialist competitive algorithm. Bulletin of Engineering Geology and
of an environmental issue of mine blasting. Engineering Computations.
the Environment, 74(3), 873–886.
Esmaeili, M., Osanloo, M., Rashidinejad, F., Aghajani Bazzazi, A., & Taji, M. (2014).
Hajihassani, M., Jahed Armaghani, D., Monjezi, M., Mohamad, E. T., & Marto, A.
Multiple regression, ANN and ANFIS models for prediction of backbreak in the
(2015). Blast-induced air and ground vibration prediction: A particle swarm
open pit blasting. Engineering with Computers, 30(4), 549–558.
optimization-based artificial neural network approach. Environmental Earth Sciences,
Esmaeili, M., Salimi, A., Drebenstedt, C., Abbaszadeh, M., & Aghajani Bazzazi, A.
74(4), 2799–2817.
(2015). Application of PCA, SVR, and ANFIS for modeling of rock fragmentation.
Halevy, A., Norvig, P., & Pereira, F. (2009). The unreasonable effectiveness of data.
Arabian Journal of Geosciences, 8(9), 6881–6893.
Fang, Q., Nguyen, H., Bui, X. N., Nguyen-Thoi, T., & Zhou, J. (2021). Modeling of rock IEEE Intelligent Systems, 24(2), 8–12.
fragmentation by firefly optimization algorithm and boosted generalized additive Han, H., Jahed Armaghani, D., Tarinejad, R., Zhou, J., & Tahir, M. (2020). Random
model. Neural Computing and Applications, 33(8), 3503–3519. forest and Bayesian network techniques for probabilistic prediction of flyrock
Fang, Q., Nguyen, H., Bui, X. N., & Tran, Q. H. (2020). Estimation of blast-induced induced by blasting in quarry sites. Natural Resources Research, 29(2), 655–667.
air overpressure in quarry mines using cubist-based genetic algorithm. Natural Harandizadeh, H., & Armaghani, D. J. (2021). Prediction of air-overpressure induced
Resources Research, 29(2), 593–607. by blasting using an ANFIS-PNN model optimized by GA. Applied Soft Computing,
Faradonbeh, R. S., Armaghani, D. J., Amnieh, H. B., & Mohamad, E. T. (2018). Predic- 99, Article 106904.
tion and minimization of blast-induced flyrock using gene expression programming Hasanipanah, M., Amnieh, H. B., Arab, H., & Zamzam, M. S. (2018). Feasibility of PSO–
and firefly algorithm. Neural Computing and Applications, 29(6), 269–281. ANFIS model to estimate rock fragmentation produced by mine blasting. Neural
Faradonbeh, R. S., Armaghani, D. J., Monjezi, M., & Mohamad, E. T. (2016). Genetic Computing and Applications, 30(4), 1015–1024.
programming and gene expression programming for flyrock assessment due to mine Hasanipanah, M., & Bakhshandeh Amnieh, H. (2021). Developing a new uncertain rule-
blasting. International Journal of Rock Mechanics and Mining Sciences, 88, 254–264. based fuzzy approach for evaluating the blast-induced backbreak. Engineering with
Faradonbeh, R. S., Hasanipanah, M., Amnieh, H. B., Armaghani, D. J., & Monjezi, M. Computers, 37(3), 1879–1893.
(2018). Development of GP and GEP models to estimate an environmental issue Hasanipanah, M., Faradonbeh, R. S., Amnieh, H. B., Armaghani, D. J., & Monjezi, M.
induced by blasting operation. Environmental Monitoring and Assessment, 190(6), (2017). Forecasting blast-induced ground vibration developing a CART model.
1–15. Engineering with Computers, 33(2), 307–316.
Faradonbeh, R. S., Jahed Armaghani, D., & Monjezi, M. (2016). Development of a new Hasanipanah, M., Faradonbeh, R. S., Armaghani, D. J., Amnieh, H. B., & Khandelwal, M.
model for predicting flyrock distance in quarry blasting: A genetic programming (2017). Development of a precise model for prediction of blast-induced flyrock
technique. Bulletin of Engineering Geology and the Environment, 75(3), 993–1006. using regression tree technique. Environmental Earth Sciences, 76(1), 1–10.
Faradonbeh, R. S., & Monjezi, M. (2017). Prediction and minimization of blast- Hasanipanah, M., Jahed Armaghani, D., Bakhshandeh Amnieh, H., Majid, M. Z. A., &
induced ground vibration using two robust meta-heuristic algorithms. Engineering Tahir, M. (2017). Application of PSO to develop a powerful equation for prediction
with Computers, 33(4), 835–851. of flyrock due to blasting. Neural Computing and Applications, 28(1), 1043–1050.
Fattahi, H., & Hasanipanah, M. (2021). Prediction of blast-induced ground vibration in Hasanipanah, M., Jahed Armaghani, D., Khamesi, H., Bakhshandeh Amnieh, H., &
a mine using relevance vector regression optimized by metaheuristic algorithms. Ghoraba, S. (2016). Several non-linear models in estimating air-overpressure
Natural Resources Research, 30(2), 1849–1863. resulting from mine blasting. Engineering with Computers, 32(3), 441–455.
Fissha, Y., Ikeda, H., Toriya, H., Adachi, T., & Kawamura, Y. (2023). Application Hasanipanah, M., Keshtegar, B., Thai, D. K., & Troung, N. T. (2020). An ANN-adaptive
of Bayesian neural network (BNN) for the prediction of blast-induced ground dynamical harmony search algorithm to approximate the flyrock resulting from
vibration. Applied Sciences, 13(5), 3128. blasting. Engineering with Computers, 1–13.
Fouladgar, N., Hasanipanah, M., & Bakhshandeh Amnieh, H. (2017). Application of Hasanipanah, M., Monjezi, M., Shahnazar, A., Armaghani, D. J., & Farazmand, A.
cuckoo search algorithm to estimate peak particle velocity in mine blasting. (2015). Feasibility of indirect determination of blast induced ground vibration
Engineering with Computers, 33(2), 181–189. based on support vector machine. Measurement, 75, 289–297.

17
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Hasanipanah, M., Naderi, R., Kashir, J., Noorani, S. A., & Zeynali Aaq Qaleh, A. (2017). Khandelwal, M., & Kankar, P. (2011). Prediction of blast-induced air overpressure using
Prediction of blast-produced ground vibration using particle swarm optimization. support vector machine. Arabian Journal of Geosciences, 4(3), 427–433.
Engineering with Computers, 33(2), 173–179. Khandelwal, M., & Monjezi, M. (2013). Prediction of backbreak in open-pit blasting op-
Hasanipanah, M., Shahnazar, A., Arab, H., Golzar, S. B., & Amiri, M. (2017). Devel- erations using the machine learning method. Rock Mechanics and Rock Engineering,
oping a new hybrid-AI model to predict blast-induced backbreak. Engineering with 46(2), 389–396.
Computers, 33(3), 349–359. Khandelwal, M., & Singh, T. (2009). Prediction of blast-induced ground vibration using
Hasanipanah, M., Shahnazar, A., Bakhshandeh Amnieh, H., & Jahed Armaghani, D. artificial neural network. International Journal of Rock Mechanics and Mining Sciences,
(2017). Prediction of air-overpressure caused by mine blasting using a new hybrid 46(7), 1214–1222.
PSO–SVR model. Engineering with Computers, 33(1), 23–31. Kitchenham, B., & Brereton, P. (2013). A systematic review of systematic review process
He, Z., Armaghani, D. J., Masoumnezhad, M., Khandelwal, M., Zhou, J., & Murlidhar, B. research in software engineering. Information and Software Technology, 55(12),
R. (2021). A combination of expert-based system and advanced decision-tree 2049–2075.
algorithms to predict air-overpressure resulting from quarry blasting. Natural Kohavi, R., et al. (1995). A study of cross-validation and bootstrap for accuracy
Resources Research, 30(2), 1889–1903. estimation and model selection. In Ijcai, vol. 14 (pp. 1137–1145). Montreal, Canada.
Holzinger, A., Biemann, C., Pattichis, C. S., & Kell, D. B. (2017). What do we need Komadja, G. C., Rana, A., Glodji, L. A., Anye, V., Jadaun, G., Onwualu, P. A., et
to build explainable AI systems for the medical domain? arXiv preprint arXiv: al. (2022). Assessing ground vibration caused by rock blasting in surface mines
1712.09923. using machine-learning approaches: A comparison of CART, SVR and MARS.
Hosseini, S., Poormirzaee, R., & Hajihassani, M. (2022). Application of reliability-based Sustainability, 14(17), 11060.
back-propagation causality-weighted neural networks to estimate air-overpressure Koopialipoor, M., Fallah, A., Armaghani, D. J., Azizi, A., & Mohamad, E. T. (2019).
due to mine blasting. Engineering Applications of Artificial Intelligence, 115, Article Three hybrid intelligent models in estimating flyrock distance resulting from
105281. blasting. Engineering with Computers, 35(1), 243–256.
Hosseini, S., Poormirzaee, R., Hajihassani, M., & Kalatehjari, R. (2022). An ANN-fuzzy Kotsiantis, S. B., Zaharakis, I., Pintelas, P., et al. (2007). Supervised machine learning:
cognitive map-based Z-number theory to predict flyrock induced by blasting in A review of classification techniques. Emerging Artificial Intelligence Applications in
open-pit mines. Rock Mechanics and Rock Engineering, 55(7), 4373–4390. Computer Engineering, 160(1), 3–24.
Hosseini, S., Pourmirzaee, R., Armaghani, D. J., & Sabri Sabri, M. M. (2023). Prediction Kulatilake, P., Hudaverdi, T., & Wu, Q. (2012). New prediction models for mean
of ground vibration due to mine blasting in a surface lead–zinc mine using machine particle size in rock blast fragmentation. Geotechnical and Geological Engineering,
learning ensemble techniques. Scientific Reports, 13(1), 6591. 30(3), 665–684.
Huang, J., Asteris, P. G., Manafi Khajeh Pasha, S., Mohammed, A. S., & Hasanipanah, M. Kumar, S., Mishra, A., & Choudhary, B. (2022). Prediction of back break in blasting
(2020). A new auto-tuning model for predicting the rock fragmentation: A cat using random decision trees. Engineering with Computers, 38(2), 1185–1191.
swarm optimization algorithm. Engineering with Computers, 1–12. Kumar, S., Mishra, A. K., & Choudhary, B. S. (2023). Estimation equations for back
Huang, J., & Xue, J. (2022). Optimization of svr functions for flyrock evaluation in break and ground vibration using genetic programming. Geotechnical and Geological
mine blasting operations. Environmental Earth Sciences, 81(17), 434. Engineering, 41(5), 3139–3149.
Hudaverdi, T. (2022). Prediction of flyrock throw distance in quarries by variable Längkvist, M., Karlsson, L., & Loutfi, A. (2014). A review of unsupervised feature
selection procedures and ANFIS modelling technique. Environmental Earth Sciences, learning and deep learning for time-series modeling. Pattern Recognition Letters, 42,
81(10), 281. 11–24.
Jahed Armaghani, D., Hajihassani, M., Marto, A., Shirani Faradonbeh, R., & Mo- Larsen, J., & Goutte, C. (1999). On optimal data split for generalization estimation and
hamad, E. T. (2015). Prediction of blast-induced air overpressure: A hybrid AI-based model selection. In Neural networks for signal processing IX: Proceedings of the 1999
predictive model. Environmental Monitoring and Assessment, 187(11), 1–13. IEEE signal processing society workshop (Cat. No. 98TH8468) (pp. 225–234). IEEE.
Jahed Armaghani, D., Hajihassani, M., Sohaei, H., Tonnizam Mohamad, E., Marto, A., Lawal, A. I., & Kwon, S. (2021). Application of artificial intelligence to rock mechan-
Motaghedi, H., et al. (2015). Neuro-fuzzy technique to predict air-overpressure ics: An overview. Journal of Rock Mechanics and Geotechnical Engineering, 13(1),
induced by blasting. Arabian Journal of Geosciences, 8(12), 10937–10950. 248–266.
Jahed Armaghani, D., Hasanipanah, M., Mahdiyar, A., Abd Majid, M. Z., Bakhshan- Lawal, A. I., Kwon, S., Hammed, O. S., & Idris, M. A. (2021). Blast-induced ground
deh Amnieh, H., & Tahir, M. (2018). Airblast prediction through a hybrid genetic vibration prediction in granite quarries: An application of gene expression pro-
algorithm-ANN model. Neural Computing and Applications, 29(9), 619–629. gramming, ANFIS, and sine cosine algorithm optimized ANN. International Journal
Jahed Armaghani, D., Hasanipanah, M., & Tonnizam Mohamad, E. (2016). A combi- of Mining Science and Technology, 31(2), 265–277.
nation of the ICA-ANN model to predict air-overpressure resulting from blasting. Li, D., Koopialipoor, M., & Armaghani, D. J. (2021). A combination of fuzzy delphi
Engineering with Computers, 32(1), 155–171. method and ANN-based models to investigate factors of flyrock induced by mine
Jahed Armaghani, D., Kumar, D., Samui, P., Hasanipanah, M., & Roy, B. (2021). A novel blasting. Natural Resources Research, 30(2), 1905–1924.
approach for forecasting of ground vibrations resulting from blasting: Modified Li, E., Yang, F., Ren, M., Zhang, X., Zhou, J., & Khandelwal, M. (2021). Prediction of
particle swarm optimization coupled extreme learning machine. Engineering with blasting mean fragment size using support vector regression combined with five
Computers, 37(4), 3221–3235. optimization algorithms. Journal of Rock Mechanics and Geotechnical Engineering,
Jain, J., Agrawal, A., & Choudhary, B. S. (2022). An advance tool to predict ground 13(6), 1380–1397.
vibration using effective blast design parameters. Current Science, 123(7), 887–894. Longjun, D., Xibing, L., Ming, X., & Qiyue, L. (2011). Comparisons of random forest and
Jamei, M., Hasanipanah, M., Karbasi, M., Ahmadianfar, I., & Taherifar, S. (2021). support vector machine for predicting blasting vibration characteristic parameters.
Prediction of flyrock induced by mine blasting using a novel kernel-based extreme Procedia Engineering, 26, 1772–1781.
learning machine. Journal of Rock Mechanics and Geotechnical Engineering, 13(6), Mahdiyar, A., Jahed Armaghani, D., Koopialipoor, M., Hedayat, A., Abdullah, A.,
1438–1451. & Yahya, K. (2020). Practical risk assessment of ground vibrations resulting
Jordan, M. I., & Mitchell, T. M. (2015). Machine learning: Trends, perspectives, and from blasting, using gene expression programming and Monte Carlo simulation
prospects. Science, 349(6245), 255–260. techniques. Applied Sciences, 10(2), 472.
Joseph, V. R. (2022). Optimal ratio for data splitting. Statistical Analysis and Data Mining: Marto, A., Hajihassani, M., Jahed Armaghani, D., Tonnizam Mohamad, E., &
The ASA Data Science Journal, 15(4), 531–538. Makhtar, A. M. (2014). A novel approach for blast-induced flyrock prediction based
Kadingdi, F. A., Ayawah, P. E., Azure, J. W., Bruno, K. A., Kaba, A. G., & Frimpong, S. on imperialist competitive algorithm and artificial neural network. The Scientific
(2022). Stacked generalization for improved prediction of ground vibration from World Journal, 2014.
blasting in open-pit mine operations. Mining, Metallurgy & Exploration, 39(6), Mehrdanesh, A., Monjezi, M., Khandelwal, M., & Bayat, P. (2021). Application of
2351–2363. various robust techniques to study and evaluate the role of effective parameters
Kapoor, S., & Narayanan, A. (2022). Leakage and the reproducibility crisis in ML-based on rock fragmentation. Engineering with Computers, 1–11.
science. arXiv preprint arXiv:2207.07048. Mehrdanesh, A., Monjezi, M., & Sayadi, A. R. (2018). Evaluation of effect of rock mass
Ke, B., Nguyen, H., Bui, X. N., & Costache, R. (2021). Estimation of ground vibration properties on fragmentation using robust techniques. Engineering with Computers,
intensity induced by mine blasting using a state-of-the-art hybrid autoencoder 34(2), 253–260.
neural network and support vector regression model. Natural Resources Research, Miao, Y., Zhang, Y., Wu, D., Li, K., Yan, X., & Lin, J. (2021). Rock fragmentation size
30(5), 3853–3864. distribution prediction and blasting parameter optimization based on the muck-pile
Keshtegar, B., Piri, J., Asnida Abdullah, R., Hasanipanah, M., Muayad Sabri Sabri, M., model. Mining, Metallurgy & Exploration, 38(2), 1071–1080.
& Nguyen Le, B. (2023). Intelligent ground vibration prediction in surface mines Mitchell, T. M. (1980a). The need for biases in learning generalizations. Citeseer.
using an efficient soft computing method based on field data. Frontiers in Public Mitchell, T. M. (1980b). The need for biases in learning generalizations. Rutgers University.
Health, 10, Article 1094771. Mohamed, M. T. (2009). Artificial neural network for prediction and control of
Khandelwal, M. (2011). Blast-induced ground vibration prediction using support vector blasting vibrations in Assiut (Egypt) limestone quarry. International Journal of Rock
machine. Engineering with Computers, 27(3), 193–200. Mechanics and Mining Sciences, 46(2), 426–431.
Khandelwal, M., Armaghani, D. J., Faradonbeh, R. S., Yellishetty, M., Majid, M. Z. A., Mohammadi, D., Mikaeil, R., & Abdollahi-Sharif, J. (2020). Implementation of an opti-
& Monjezi, M. (2017). Classification and regression tree technique in estimating mized binary classification by GMDH-type neural network algorithm for predicting
peak particle velocity caused by blasting. Engineering with Computers, 33(1), 45–53. the blast produced ground vibration. Expert Systems, 37(5), Article e12563.

18
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Mohammadnejad, M., Gholami, R., Sereshki, F., & Jamshidi, A. (2013). A new Nguyen, H., Bui, X. N., Tran, Q. H., & Mai, N.-L. (2019). A new soft computing
methodology to predict backbreak in blasting operation. International Journal of model for estimating and controlling blast-produced ground vibration based on
Rock Mechanics and Mining Sciences, 60, 75–81. hierarchical K-means clustering and cubist algorithms. Applied Soft Computing, 77,
Mojtahedi, S. F. F., Ebtehaj, I., Hasanipanah, M., Bonakdari, H., & Amnieh, H. B. 376–386.
(2019). Proposing a novel hybrid intelligent model for the simulation of particle Nguyen, H., Bui, X. N., Tran, Q. H., & Moayedi, H. (2019). Predicting blast-induced
size distribution resulting from blasting. Engineering with Computers, 35(1), 47–56. peak particle velocity using BGAMs, ANN and SVM: A case study at the Nui Beo
Monjezi, M., Ahmadi, Z., Varjani, A. Y., & Khandelwal, M. (2013). Backbreak prediction open-pit coal mine in Vietnam. Environmental Earth Sciences, 78(15), 1–14.
in the chadormalu iron mine using artificial neural network. Neural Computing and Nguyen, H., Bui, X. N., Tran, Q. H., Nguyen, D. A., Hoa, L. T. T., Le, Q. T., et al.
Applications, 23(3), 1101–1107. (2021). Predicting blast-induced ground vibration in open-pit mines using different
Monjezi, M., Amini Khoshalan, H., & Yazdian Varjani, A. (2012). Prediction of flyrock nature-inspired optimization algorithms and deep neural network. Natural Resources
and backbreak in open pit blasting operation: A neuro-genetic approach. Arabian Research, 30(6), 4695–4717.
Journal of Geosciences, 5(3), 441–448. Nguyen, H., Bui, X. N., Tran, Q. H., Van Hoa, P., Nguyen, D. A., Hoa, L. T. T.,
Monjezi, M., Amiri, H., Farrokhi, A., & Goshtasbi, K. (2010). Prediction of rock et al. (2020). A comparative study of empirical and ensemble machine learning
fragmentation due to blasting in Sarcheshmeh copper mine using artificial neural algorithms in predicting air over-pressure in open-pit coal mine. Acta Geophysica,
networks. Geotechnical and Geological Engineering, 28(4), 423–430. 68(2), 325–336.
Monjezi, M., Baghestani, M., Shirani Faradonbeh, R., Pourghasemi Saghand, M., & Nguyen, H., Choi, Y., Bui, X. N., & Nguyen Thoi, T. (2019). Predicting blast-induced
Jahed Armaghani, D. (2016). Modification and prediction of blast-induced ground ground vibration in open-pit mines using vibration sensors and support vector
vibrations based on both empirical and computational techniques. Engineering with regression-based optimization algorithms. Sensors, 20(1), 132.
Computers, 32(4), 717–728. Nguyen, H., Drebenstedt, C., Bui, X. N., & Bui, D. T. (2020). Prediction of blast-induced
Monjezi, M., Bahrami, A., & Varjani, A. Y. (2010). Simultaneous prediction of ground vibration in an open-pit mine by a novel hybrid model based on clustering
fragmentation and flyrock in blasting operation using artificial neural networks. and artificial neural network. Natural Resources Research, 29(2), 691–709.
International Journal of Rock Mechanics and Mining Sciences, 47(3), 476–480. Nikafshan Rad, H., Bakhshayeshi, I., Wan Jusoh, W. A., Tahir, M., & Foong, L. K.
Monjezi, M., Bahrami, A., Varjani, A. Y., & Sayadi, A. R. (2011). Prediction and (2020). Prediction of flyrock in mine blasting: A new computational intelligence
controlling of flyrock in blasting operation using artificial neural network. Arabian approach. Natural Resources Research, 29(2), 609–623.
Journal of Geosciences, 4(3), 421–425. Ohadi, B., Sun, X., Esmaieli, K., & Consens, M. P. (2020). Predicting blast-induced
Monjezi, M., & Dehghani, H. (2008). Evaluation of effect of blasting pattern parameters outcomes using random forest models of multi-year blasting data from an open pit
on back break using neural networks. International Journal of Rock Mechanics and mine. Bulletin of Engineering Geology and the Environment, 79(1), 329–343.
Mining Sciences, 45(8), 1446–1453. Peng, K., Zeng, J., Armaghani, D. J., Hasanipanah, M., & Chen, Q. (2021). A novel com-
Monjezi, M., Dehghani, H., Shakeri, J., & Mehrdanesh, A. (2021). Optimization of bination of gradient boosted tree and optimized ANN models for forecasting ground
prediction of flyrock using linear multivariate regression (LMR) and gene expression vibration due to quarry blasting. Natural Resources Research, 30(6), 4657–4671.
programming (GEP)—Topal Novin mine, Iran. Arabian Journal of Geosciences,
Qi, C. c. (2020). Big data management in the mining industry. International Journal of
14(15), 1–12.
Minerals, Metallurgy and Materials, 27(2), 131–139.
Monjezi, M., Khoshalan, H. A., & Varjani, A. Y. (2011). Optimization of open pit blast
Qiu, Y., Zhou, J., Khandelwal, M., Yang, H., Yang, P., & Li, C. (2021). Performance
parameters using genetic algorithm. International Journal of Rock Mechanics and
evaluation of hybrid WOA-XGBoost, GWO-XGBoost and BO-XGBoost models to
Mining Sciences, 48(5), 864–869.
predict blast-induced ground vibration. Engineering with Computers, 1–18.
Monjezi, M., Mehrdanesh, A., Malek, A., & Khandelwal, M. (2013). Evaluation of
Rad, H. N., Hasanipanah, M., Rezaei, M., & Eghlim, A. L. (2018). Developing a least
effect of blast design parameters on flyrock using artificial neural networks. Neural
squares support vector machine for estimating the blast-induced flyrock. Engineering
Computing and Applications, 23(2), 349–356.
with Computers, 34(4), 709–717.
Monjezi, M., Mohamadi, H. A., Barati, B., & Khandelwal, M. (2014). Application of soft
Rahm, E., Do, H. H., et al. (2000). Data cleaning: Problems and current approaches.
computing in predicting rock fragmentation to reduce environmental blasting side
IEEE Data Engineering Bulletin, 23(4), 3–13.
effects. Arabian Journal of Geosciences, 7(2), 505–511.
Ramesh Murlidhar, B., Yazdani Bejarbaneh, B., Jahed Armaghani, D., Mohammed, A.
Murlidhar, B. R., Kumar, D., Jahed Armaghani, D., Mohamad, E. T., Roy, B., & Pham, B.
S., & Tonnizam Mohamad, E. (2021). Application of tree-based predictive models
T. (2020). A novel intelligent ELM-BBO technique for predicting distance of mine
to forecast air overpressure induced by mine blasting. Natural Resources Research,
blasting-induced flyrock. Natural Resources Research, 29(6), 4103–4120.
30(2), 1865–1887.
Murlidhar, B. R., Nguyen, H., Rostami, J., Bui, X., Armaghani, D. J., Ragam, P., et
Rezaeineshat, A., Monjezi, M., Mehrdanesh, A., & Khandelwal, M. (2020). Optimization
al. (2021). Prediction of flyrock distance induced by mine blasting using a novel
of blasting design in open pit limestone mines with the aim of reducing ground
Harris Hawks optimization-based multi-layer perceptron neural network. Journal of
vibration using robust techniques. Geomechanics and Geophysics for Geo-Energy and
Rock Mechanics and Geotechnical Engineering, 13(6), 1413–1427.
Geo-Resources, 6(2), 1–14.
Nguyen, H., & Bui, X. N. (2019). Predicting blast-induced air overpressure: a robust
Rogers, W. P., Kahraman, M. M., Drews, F. A., Powell, K., Haight, J. M., Wang, Y.,
artificial intelligence system based on artificial neural networks and random forest.
et al. (2019). Automation in the mining industry: Review of technology, systems,
Natural Resources Research, 28(3), 893–907.
Nguyen, H., & Bui, X. N. (2020). Soft computing models for predicting blast-induced human factors, and political risk. Mining, Metallurgy & Exploration, 36(4), 607–631.
air over-pressure: A novel artificial intelligence approach. Applied Soft Computing, Rosales-Huamani, J. A., Perez-Alvarado, R. S., Rojas-Villanueva, U., & Castillo-
92, Article 106292. Sequera, J. L. (2020). Design of a predictive model of rock breakage by blasting
Nguyen, H., Bui, X. N., Bui, H.-B., & Cuong, D. T. (2019). Developing an XGBoost using artificial neural networks. Symmetry, 12(9), 1405.
model to predict blast-induced peak particle velocity in an open-pit mine: A case Sadeghi, F., Monjezi, M., & Jahed Armaghani, D. (2020). Evaluation and optimization
study. Acta Geophysica, 67(2), 477–490. of prediction of toe that arises from mine blasting operation using various soft
Nguyen, H., Bui, X. N., Bui, H. B., & Mai, N. L. (2020). A comparative study of artificial computing techniques. Natural Resources Research, 29(2), 887–903.
neural networks in predicting blast-induced air-blast overpressure at Deo Nai Saghatforoush, A., Monjezi, M., Shirani Faradonbeh, R., & Jahed Armaghani, D.
open-pit coal mine, Vietnam. Neural Computing and Applications, 32(8), 3939–3955. (2016). Combination of neural network and ant colony optimization algorithms
Nguyen, H., Bui, X. N., Choi, Y., Lee, C. W., & Armaghani, D. J. (2021). A novel for prediction and optimization of flyrock and back-break induced by blasting.
combination of whale optimization algorithm and support vector machine with Engineering with Computers, 32(2), 255–266.
different kernel functions for prediction of blasting-induced fly-rock in quarry Sayadi, A., Monjezi, M., Talebi, N., & Khandelwal, M. (2013). A comparative study
mines. Natural Resources Research, 30(1), 191–207. on the application of various artificial neural networks to simultaneous prediction
Nguyen, H., Bui, X. N., & Moayedi, H. (2019). A comparison of advanced computational of rock fragmentation and backbreak. Journal of Rock Mechanics and Geotechnical
models and experimental techniques in predicting blast-induced ground vibration Engineering, 5(4), 318–324.
in open-pit coal mine. Acta Geophysica, 67(4), 1025–1037. Sayevand, K., & Arab, H. (2019). A fresh view on particle swarm optimization to
Nguyen, H., Bui, X. N., Nguyen-Thoi, T., Ragam, P., & Moayedi, H. (2019). Toward develop a precise model for predicting rock fragmentation. Engineering Computations.
a state-of-the-art of fly-rock prediction technology in open-pit mines using EANNs Sayevand, K., Arab, H., & Golzar, S. B. (2018). Development of imperialist competitive
model. Applied Sciences, 9(21), 4554. algorithm in predicting the particle size distribution after mine blasting. Engineering
Nguyen, H., Bui, X. N., & Topal, E. (2023a). Enhancing predictions of blast-induced with Computers, 34(2), 329–338.
ground vibration in open-pit mines: Comparing swarm-based optimization algo- Shahmoradi, J., Talebi, E., Roghanchi, P., & Hassanalian, M. (2020). A comprehensive
rithms to optimize self-organizing neural networks. International Journal of Coal review of applications of drone technology in the mining industry. Drones, 4(3),
Geology, Article 104294. 34.
Nguyen, H., Bui, X. N., & Topal, E. (2023b). Reliability and availability artificial Shams, S., Monjezi, M., Majd, V. J., & Armaghani, D. J. (2015). Application of fuzzy
intelligence models for predicting blast-induced ground vibration intensity in open- inference system for prediction of rock fragmentation induced by blasting. Arabian
pit mines to ensure the safety of the surroundings. Reliability Engineering & System Journal of Geosciences, 8(12), 10819–10832.
Safety, 231, Article 109032. Shang, Y., Nguyen, H., Bui, X. N., Tran, Q. H., & Moayedi, H. (2020). A novel artificial
Nguyen, H., Bui, X. N., & Tran, Q. H. (2021). Estimating air over-pressure resulting intelligence approach to predict blast-induced ground vibration in open-pit mines
from blasting in quarries based on a novel ensemble model (GLMNETs–MLPNN). based on the firefly algorithm and artificial neural network. Natural Resources
Natural Resources Research, 30(3), 2629–2646. Research, 29(2), 723–737.

19
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Sharma, M., Agrawal, H., & Choudhary, B. (2022). Multivariate regression and genetic Yari, M., Bagherpour, R., Jamali, S., & Shamsi, R. (2016). Development of a novel
programming for prediction of backbreak in open-pit blasting. Neural Computing flyrock distance prediction model using BPNN for providing blasting operation
and Applications, 34(3), 2103–2114. safety. Neural Computing and Applications, 27(3), 699–706.
Sharma, M., Choudhary, B. S., & Agrawal, H. (2022). Prediction of backbreak in Yari, M., He, B., Armaghani, D. J., Abbasi, P., & Mohamad, E. T. (2023). A novel ensem-
hot strata/fiery seam of open-pit coal mine by decision tree and random forest ble machine learning model to predict mine blasting–induced rock fragmentation.
algorithm. Arabian Journal of Geosciences, 15(15), 1337. Bulletin of Engineering Geology and the Environment, 82(5), 187.
Shi, X. z., Jian, Z., Wu, B. b., Huang, D., & Wei, W. (2012). Support vector machines Ye, J., Dalle, J., Nezami, R., Hasanipanah, M., & Armaghani, D. J. (2020). Stochas-
approach to mean particle size of rock fragmentation due to bench blasting
tic fractal search-tuned ANFIS model to predict blast-induced air overpressure.
prediction. Transactions of the Nonferrous Metals Society of China, 22(2), 432–441.
Engineering with Computers, 1–15.
Shirani Faradonbeh, R., Jahed Armaghani, D., Abd Majid, M., Md Tahir, M.,
Ye, J., Koopialipoor, M., Zhou, J., Armaghani, D. J., & He, X. (2021). A novel
Ramesh Murlidhar, B., Monjezi, M., et al. (2016). Prediction of ground vibration
combination of tree-based modeling and Monte Carlo simulation for assessing
due to quarry blasting based on gene expression programming: A new model for
risk levels of flyrock induced by mine blasting. Natural Resources Research, 30(1),
peak particle velocity prediction. International Journal of Environmental Science and
225–243.
Technology, 13(6), 1453–1464.
Shirani Faradonbeh, R., Monjezi, M., & Jahed Armaghani, D. (2016). Genetic program- Yu, C., Koopialipoor, M., Murlidhar, B. R., Mohammed, A. S., Armaghani, D. J.,
ing and non-linear multiple regression techniques to predict backbreak in blasting Mohamad, E. T., et al. (2021). Optimal ELM–Harris Hawks optimization and ELM–
operation. Engineering with Computers, 32(1), 123–133. Grasshopper optimization models to forecast peak particle velocity resulting from
Singh, T., & Singh, V. (2005). An intelligent approach to prediction and control ground mine blasting. Natural Resources Research, 30(3), 2647–2662.
vibration in mines. Geotechnical & Geological Engineering, 23(3), 249–262. Yu, Q., Monjezi, M., Mohammed, A. S., Dehghani, H., Armaghani, D. J., & Ulrikh, D.
Stulp, F., & Sigaud, O. (2015). Many regression algorithms, one unified model: A V. (2021). Optimized support vector machines combined with evolutionary random
review. Neural Networks, 69, 60–79. forest for prediction of back-break caused by blasting operation. Sustainability,
Sun, S., Cao, Z., Zhu, H., & Zhao, J. (2019). A survey of optimization methods from a 13(22), 12797.
machine learning perspective. IEEE Transactions on Cybernetics, 50(8), 3668–3681. Yu, Z., Shi, X., Miao, X., Zhou, J., Khandelwal, M., Chen, X., et al. (2021). Intelligent
Sutton, R. S., & Barto, A. G. (2018). Reinforcement learning: An introduction. MIT Press. modeling of blast-induced rock movement prediction using dimensional analysis
Taheri, K., Hasanipanah, M., Golzar, S. B., & Majid, M. Z. A. (2017). A hybrid artificial and optimized artificial neural network technique. International Journal of Rock
bee colony algorithm-artificial neural network for forecasting the blast-produced Mechanics and Mining Sciences, 143, Article 104794.
ground vibration. Engineering with Computers, 33(3), 689–700.
Yu, Z., Shi, X., Zhou, J., Chen, X., Miao, X., Teng, B., et al. (2020). Prediction of
Temeng, V. A., Arthur, C. K., & Ziggah, Y. Y. (2022). Suitability assessment of different
blast-induced rock movement during bench blasting: Use of gray wolf optimizer
vector machine regression techniques for blast-induced ground vibration prediction
and support vector regression. Natural Resources Research, 29(2), 843–865.
in Ghana. Modeling Earth Systems and Environment, 8(1), 897–909.
Yu, Z., Shi, X., Zhou, J., Chen, X., & Qiu, X. (2020). Effective assessment of blast-
Temeng, V. A., Ziggah, Y. Y., & Arthur, C. K. (2020). A novel artificial intelligent model
for predicting air overpressure using brain inspired emotional neural network. induced ground vibration using an optimized random forest model based on a
International Journal of Mining Science and Technology, 30(5), 683–689. Harris Hawks optimization algorithm. Applied Sciences, 10(4), 1403.
Temeng, V. A., Ziggah, Y. Y., & Arthur, C. K. (2021). Blast-induced noise level Yu, Z., Shi, X., Zhou, J., Gou, Y., Huo, X., Zhang, J., et al. (2020). A new multikernel
prediction model based on brain inspired emotional neural network. Journal of relevance vector machine based on the HPSOGWO algorithm for predicting and
Sustainable Mining, 20. controlling blast-induced ground vibration. Engineering with Computers, 1–16.
Teney, D., Peyrard, M., & Abbasnejad, E. (2022). Predicting is not understanding: Yu, Z., Shi, X., Zhou, J., Rao, D., Chen, X., Dong, W., et al. (2021). Feasibility of the
Recognizing and addressing underspecification in machine learning. In Computer indirect determination of blast-induced rock movement based on three new hybrid
vision–ECCV 2022: 17th European conference, Tel Aviv, Israel, October 23–27, 2022, intelligent models. Engineering with Computers, 37(2), 991–1006.
Proceedings, Part XXIII (pp. 458–476). Springer. Zangoei, A., Monjezi, M., Armaghani, D. J., Mehrdanesh, A., & Ahmadian, S. (2022).
Tonnizam Mohamad, E., Jahed Armaghani, D., Hasanipanah, M., Murlidhar, B. R., & Prediction and optimization of flyrock and oversize boulder induced by mine
Alel, M. N. A. (2016). Estimation of air-overpressure produced by blasting operation blasting using artificial intelligence techniques. Environmental Earth Sciences, 81(13),
through a neuro-genetic technique. Environmental Earth Sciences, 75(2), 1–15. 1–13.
Trivedi, R., Singh, T., & Gupta, N. (2015). Prediction of blast-induced flyrock in Zeng, J., Jamei, M., Nait Amar, M., Hasanipanah, M., & Bayat, P. (2022). A novel
opencast mines using ANN and ANFIS. Geotechnical and Geological Engineering, solution for simulating air overpressure resulting from blasting using an efficient
33(4), 875–891.
cascaded forward neural network. Engineering with Computers, 38(3), 2069–2081.
Van Engelen, J. E., & Hoos, H. H. (2020). A survey on semi-supervised learning. Machine
Zeng, J., Roussis, P. C., Mohammed, A. S., Maraveas, C., Fatemi, S. A., Armaghani, D.
Learning, 109(2), 373–440.
J., et al. (2021). Prediction of peak particle velocity caused by blasting through
Van der Walt, J., & Spiteri, W. (2020). A critical analysis of recent research into the
the combinations of boosted-CHAID and SVM models with various kernels. Applied
prediction of flyrock and related issues resulting from surface blasting activities.
Journal of the Southern African Institute of Mining and Metallurgy, 120(12), 701–714. Sciences, 11(8), 3705.
Wang, X., Hosseini, S., Jahed Armaghani, D., & Tonnizam Mohamad, E. (2023). Zhang, S., Bui, X. N., Trung, N. T., Nguyen, H., & Bui, H. B. (2020). Prediction of rock
Data-driven optimized artificial neural network technique for prediction of flyrock size distribution in mine bench blasting using a novel ant colony optimization-based
induced by boulder blasting. Mathematics, 11(10), 2358. boosted regression tree technique. Natural Resources Research, 29(2), 867–886.
Wolpert, D. H. (1996). The lack of a priori distinctions between learning algorithms. Zhang, R., Li, Y., Gui, Y., & Zhou, J. (2022). Prediction of blasting induced air-
Neural Computation, 8(7), 1341–1390. overpressure using a radial basis function network with an additional hidden layer.
Xie, C., Nguyen, H., Bui, X. N., Choi, Y., Zhou, J., & Nguyen Trang, T. (2021). Predicting Applied Soft Computing, 127, Article 109343.
rock size distribution in mine blasting using various novel soft computing models Zhang, H., Zhou, J., Jahed Armaghani, D., Tahir, M., Pham, B. T., & Huynh, V. V.
based on meta-heuristics and machine learning algorithms. Geoscience Frontiers, (2020). A combination of feature selection and random forest techniques to solve
12(3), Article 101108. a problem related to blast-induced ground vibration. Applied Sciences, 10(3), 869.
Xin, D., Wu, E. Y., Lee, D. J. L., Salehi, N., & Parameswaran, A. (2021). Whither Zhongya, Z., & Xiaoguang, J. (2018). Prediction of peak velocity of blasting vibration
automl? Understanding the role of automation in machine learning workflows. In based on artificial neural network optimized by dimensionality reduction of
Proceedings of the 2021 CHI conference on human factors in computing systems (pp. FA-MIV. Mathematical Problems in Engineering, 2018.
1–16).
Zhou, J., Aghili, N., Ghaleini, E. N., Bui, D. T., Tahir, M., & Koopialipoor, M. (2020).
Xu, Y., & Goodacre, R. (2018). On splitting training and validation set: A comparative
A Monte Carlo simulation approach for effective assessment of flyrock based on
study of cross-validation, bootstrap and systematic sampling for estimating the
intelligent system of neural network. Engineering with Computers, 36(2), 713–723.
generalization performance of supervised learning. Journal of Analysis and Testing,
Zhou, X., Armaghani, D. J., Ye, J., Khari, M., & Motahari, M. R. (2021). Hybridization
2(3), 249–262.
Yan, Y., Hou, X., & Fei, H. (2020). Review of predicting the blast-induced ground of parametric and non-parametric techniques to predict air over-pressure induced
vibrations to reduce impacts on ambient urban communities. Journal of Cleaner by quarry blasting. Natural Resources Research, 30(1), 209–224.
Production, 260, Article 121135. Zhou, J., Asteris, P. G., Armaghani, D. J., & Pham, B. T. (2020). Prediction of ground
Yang, H., Hasanipanah, M., Tahir, M., & Bui, D. T. (2020). Intelligent prediction of vibration induced by blasting operations through the use of the Bayesian network
blasting-induced ground vibration using ANFIS optimized by GA and PSO. Natural and random forest models. Soil Dynamics and Earthquake Engineering, 139, Article
Resources Research, 29(2), 739–750. 106390.
Yang, H., Nikafshan Rad, H., Hasanipanah, M., Bakhshandeh Amnieh, H., & Nekouie, A. Zhou, J., Dai, Y., Khandelwal, M., Monjezi, M., Yu, Z., & Qiu, Y. (2021). Performance
(2020). Prediction of vibration velocity generated in mine blasting using support of hybrid SCA-RF and HHO-RF models for predicting backbreak in open-pit mine
vector regression improved by optimization algorithms. Natural Resources Research, blasting operations. Natural Resources Research, 30(6), 4753–4771.
29(2), 807–830. Zhou, J., Koopialipoor, M., Murlidhar, B. R., Fatemi, S. A., Tahir, M., Jahed Ar-
Yari, M., Armaghani, D. J., Maraveas, C., Ejlali, A. N., Mohamad, E. T., & Asteris, P. maghani, D., et al. (2020). Use of intelligent methods to design effective pattern
G. (2023). Several tree-based solutions for predicting flyrock distance due to mine parameters of mine blasting to minimize flyrock distance. Natural Resources
blasting. Applied Sciences, 13(3), 1345. Research, 29(2), 625–639.

20
V. Munagala et al. Machine Learning with Applications 15 (2024) 100517

Zhou, J., Li, C., Arslan, C. A., Hasanipanah, M., & Bakhshandeh Amnieh, H. (2021). Zhou, J., Qiu, Y., Khandelwal, M., Zhu, S., & Zhang, X. (2021). Developing a hybrid
Performance evaluation of hybrid FFA-ANFIS and GA-ANFIS models to predict model of Jaya algorithm-based extreme gradient boosting machine to estimate
particle size distribution of a muck-pile after blasting. Engineering with Computers, blast-induced ground vibrations. International Journal of Rock Mechanics and Mining
37(1), 265–274. Sciences, 145, Article 104856.
Zhou, J., Li, C., Koopialipoor, M., Jahed Armaghani, D., & Thai Pham, B. (2021). Zhu, X. J. (2005). Semi-supervised learning literature survey. University of Wisconsin-
Development of a new methodology for estimating the amount of PPV in surface Madison Department of Computer Sciences.
mines based on prediction and probabilistic models (GEP-MC). International Journal Zhu, W., Rad, H. N., & Hasanipanah, M. (2021). A chaos recurrent ANFIS optimized by
of Mining, Reclamation and Environment, 35(1), 48–68. PSO to predict ground vibration generated in rock blasting. Applied Soft Computing,
Zhou, J., Nekouie, A., Arslan, C. A., Pham, B. T., & Hasanipanah, M. (2020). Novel 108, Article 107434.
approach for forecasting the blast-induced AOp using a hybrid fuzzy system and Ziggah, Y. Y., Temeng, V. A., & Arthur, C. K. (2023). A new synergetic model of neigh-
firefly algorithm. Engineering with Computers, 36(2), 703–712. bourhood component analysis and artificial intelligence method for blast-induced
noise prediction. Modeling Earth Systems and Environment, 1–20.

21

You might also like