Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

Safety Science 144 (2021) 105481

Contents lists available at ScienceDirect

Safety Science
journal homepage: www.elsevier.com/locate/safety

Discovery of new safety knowledge from mining large injury dataset


in construction
Xiaoxiao Xu a, Patrick X.W. Zou b, c, *
a
School of Civil Engineering, Nanjing Forestry University, Nanjing, Jiangsu 210037, China
b
School of Economics and Management, Chang’an University, Xi’an, Shaanxi 710064, China
c
Engineering Research Center of Highway Infrastructure Digitalization, Ministry of Education, Chang’an University, Xi’an, Shaanxi 710064, China

A R T I C L E I N F O A B S T R A C T

Keywords: Safety in construction continues to remain as the 3rd most serious occupational problem worldwide. Previous
Construction safety research mainly used questionnaire surveys or interviews to obtain “cross-sectional” data, which may be difficult
Injury to extract patterns and gain new knowledge for improving safety performance. This research aims to discover the
Incident
characteristics, patterns, mechanisms and knowledge from mining a large injury dataset, and to develop stra­
Knowledge discovery
Visualisation
tegies for mitigating safety risks, and reducing accidents and injuries so that to improve construction safety
Association rule mining performance. A large set of injury and incident data recorded by construction companies over a period of 11
years was used in this research. Statistical analysis, visualisation analysis and association rule mining (ARM)
methods were applied to mine and analyse this large dataset. “Injury time”, “injured body locations, organs and
systems”, “age distribution of injured”, “causes, nature and relationships of injury, and injured body parts”
“changing trends of incidents”, were analysed in depth to discover new safety knowledge. Based on the
knowledge and mechanism discovered, this study proposed five strategies for improving construction safety
performance covering safety skills, safety awareness, safety protection and emergency response. This research
would be valuable to both researchers and practitioners. Researchers would benefit from gaining a deeper un­
derstanding of the characteristics of construction safety injuries, application of data mining methods, and
relevant future research directions, while practitioners would benefit by learning about the strategies proposed to
mitigate injury risks and prevent incidents.

1. Introduction et al., 2020). At a global level, almost one person is killed every 5-mi­
nutes on construction sites due to bad working conditions and/or un­
The construction industry is considered as one of the most hazardous safe behaviour (Chiang et al., 2018). Construction injury will interrupt
industries around the world (Guo et al., 2021; Zou and Sunindijo, 2015) the orderly progress of construction work, and impose a considerable
and it is characterised by compacted workforces, complex working en­ economic burden on employers (Kakhki et al., 2019; Sarkar et al., 2020).
vironments and construction methods, temporary workplaces and or­ Moreover, an injury will affect workers’ life with chronic pain, psy­
ganizations, and diverse work tasks and construction machines and chological suffering, loss of income and extensive medical expenses
equipment (Jin et al., 2019; Kines, 2002). Construction workers are (Sánchez et al., 2011). These sufferings could extend to the workers’
often required to use electrical powerful machines and equipment, work family, community and the entire society. Therefore, the need for the
at heights, and carry heavy goods and items as well as working outdoor construction industry to reduce occupational injury and improve occu­
under high temperatures and tough conditions. As a result, the con­ pational safety is a matter of pressing urgency (Choi et al., 2020).
struction industry suffers more risks of occupational injuries than other This research aims to discover the characteristics, patterns, mecha­
industries (Fang et al., 2016; Larsson and Field, 2002). According to nisms and knowledge from mining a large injury dataset, and to develop
statistics, the construction industry accounts for approximately 7% of strategies for mitigating safety risks, and reducing accidents and injuries
employment but up to 20% of the world’s occupational fatalities (Choi to improve construction safety performance. Statistical analysis, data

* Corresponding author at: School of Economics and Management & Engineering Research Center of Highway Infrastructure Digitalization, Ministry of Education,
Chang’an University, Middle Section, Nan’er Huan Road, Xi’an, Shaanxi 710064, China.
E-mail address: pxwzou@outlook.com (P.X.W. Zou).

https://doi.org/10.1016/j.ssci.2021.105481
Received 26 March 2021; Received in revised form 30 July 2021; Accepted 1 September 2021
Available online 9 September 2021
0925-7535/© 2021 Elsevier Ltd. All rights reserved.
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

visualisation and association rule mining methods are applied in this analysis, machine learning and nature language processing should be
research. The novelty of this study is that it provides a “panorama” and applied. Zhang and Zhang (2003) held the view that data mining could
also a deeper and more detailed understanding of construction injury identify potential useful patterns in the large volume and high dimen­
with the consideration of time of the injury accidents, bodily location of sionality of construction injury databases. The main data mining
injury, age of the injured, as well as the nature and formation mecha­ methods in use include classification and regression tree (Cheng et al.,
nism of injury. The results of this study are crucial to construction safety 2012), association rule mining (Liao and Perng, 2008), decision tree
management as they can serve as a solid foundation for developing more modelling (Amiri et al., 2016). For instance, Tixier et al. (2017) applied
effective and efficient safety management strategies. The remainder of data mining techniques to discover attribute combinations that
this paper is structured as follows. Pursuant to this introductory section contribute to injuries; Mistikoglu et al. (2015) employed decision tree
is literature revew (Section 2). A detailed description of the research techniques to extract rules that shows the association in roofer fall
methods, including description of the dataset, data preparation and data accidents.
analysis methods and processes is presented in Section 3. Section 4 Regarding statistical analysis, Isaac and Edrei (2016) developed a
shows the results of the mining and discovery of injury database. Section statistical model that uses real-time tracking data to proactive alerts in
5 proposes five strategies for improving construction safety perfor­ cases of increasing risk exposure. Lander et al. (2016) analysed 23,464
mance, followed by study implications and contributions (Section 6). work-related injuries spanned over 30 years from 1980 to 2010 to
Section 7 concludes this study, with main findings and future research investigate injury trends according to age, severity, work activity and
directions. business cycle by using linear regression analysis. The annual incidences
were calculated and employment levels were used as an indicator of
2. Literature review fluctuations in the business cycle. Their main findings are the number of
injuries in the construction sector has not changed significantly in the
2.1. Overview of construction injury research past 30 years, and young workers have a higher injury level than old
workers.
The importance of undertaking research in construction safety is Machine learning is mainly applied to predict and analyse injury
reflected in the ASCE Journal of Construction Engineering and Management severity (Kakhki et al., 2019; Sarkar et al., 2020). Furthermore, machine
2017 annual publications in that 28% of the whole year publications learning is also employed to predict injury type, energy type, body part
were focused on construction safety (Sherratt and Leicht, 2020). Previ­ (Kang and Ryu, 2019; Tixier et al., 2016a). For example, Ayhan and
ous studies have attempted to explore the nature of construction injury, Tokdemir (2020) predicted the outcomes of construction incidents using
such as the associated factors of construction injuries (Berhanu et al., Latent Class Clustering Analysis and Artificial Neural Networks; Choi
2019; Gebremeskel and Yimer, 2019), precursors and outcomes of in­ et al. (2020) identified the potential risk of fatality accidents at con­
juries (Baker et al., 2020; Tixier et al., 2016b), prediction of construction struction sites using machine learning. It should be noted that most
injuries (Tixier et al., 2016a), human skills and training for safety (Zou injury and accident prediction models are viewed in terms of inputs and
and Sunindijo, 2015). There have been three distinctive phases of outputs without any knowledge of their internal workings, and thus do
research in construction safety, including traditional phase, transitional not facilitate the analysis of the mechanisms of safety incidents.
phase and transformational phase (Finneran and Gibb, 2013; Zou and In the construction industry, injury data are mainly in the form of
Sunindijo, 2015). In their book “Strategic Safety Management in Con­ summary reports. Therefore, some studies applied natural language
struction and Engineering”, Zou and Sunindijo (2015) stated that the processing to analyse construction injury reports. For example, Zhang
fourth phase would be the research on digital technology and data sci­ et al. (2019) proposed an ensemble model to extract and classify con­
ence enabled management of construction safety. However, it is not easy struction accident causes from construction accident reports. Kim and
for researchers to obtain construction injury and incident data although Chi (2019) developed a knowledge management system for construction
the amount of data being recorded is increasing at an astonishing rate accident cases by using natural language processing. This method could
(Kirk et al., 2016). This is partly because of the sensitivity of the inci­ save significant time and effort to retrieve relevant knowledge. In gen­
dent/injury data. Coupling with this, more often than not, many con­ eral, the amount of data involved in the existing studies is small, and the
struction companies did not pay sufficient attention to the safety of their information on injured body parts, causes of injury, nature of injury is
workers due to conflict with other priorities such as cost, time and not detailed. These may affect the comprehensiveness, accuracy and
quality (Ajayi et al., 2021; Hallowell et al., 2011; Panwar and Jha, validity of different models for injury analysis.
2021).
2.3. Summary and point of departure
2.2. Construction safety research methods
Although previous studies have investigated the nature of injury in
Due to the limited access to the sources of real cases and data, most of construction, there is still a lack of in-depth understanding of the char­
the previous studies usually did not depend on empirical data (Ayhan acteristics and formation mechanisms of injury. There are mainly three
and Tokdemir, 2020), instead, interview and questionnaire surveys are limitations in previous studies. First, most studies focused on the pre­
usually applied for data collection. For instance, Lette et al. (2018) diction of construction injuries. Although proposed models predict
evaluated the prevalence of injury and related factors among building injury information more accurately, most of them are “black box”
construction workers based on structured and pre-tested questionnaires, models. Researchers and readers may have no idea about how and why
and Tadesse and Israel (2016) explored prevalence and factors associ­ the model works as it works, and thus they cannot explain the formation
ated with occupational injuries with similar data collection methods. mechanisms of injury. Second, construction injury is a complex problem.
Khashaba et al. (2018) applied interviews and questionnaires to deter­ Only sufficiently large longitudinal sample size of injury cases can
mine the association between non-fatal occupational injuries among provide a comprehensive understanding of injury accidents. Influenced
construction workers and risk factors. However, the data collected in by the quality of the data, the existing studies may have difficulty in
these ways may deviate from reality, because social desirability bias extracting construction injury formation mechanism and patterns and
may exist in self-reported studies (Xu et al., 2018a). For example, hence the safety knowledge. Third, most of the existing studies applied a
Tadesse and Israel (2016) pointed out that interviewees or respondents single approach to the analysis of injury data, such as statistical analysis.
are more likely to report more socially acceptable responses than their This often results in difficulties in systematically mining the hidden
actual practice. knowledge and patterns of injury data. To overcome the aforementioned
To the extent that real data are available, data mining, statistical limitations, as a point of departure, in this current research, a

2
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

Fig. 1. Database for construction injury details.

combination of statistical analysis, data visualisation and association 3.2. Data preparation
rule analysis methods is used to uncover injury characteristics, patterns,
mechanisms and knowledge from a large set of injury data with 19 Before data analysis and data visualization, data preparation is
injured body parts, 11 injured organs and 8 injured systems, 28 causes of required to ensure data quality. Three rounds of data preparation were
injuries and 16 nature of injury. carried out. In the first round of data preparation, all unrelated data are
deleted. This study focuses on construction injury, and thus non-injury-
3. Research methods and processes related data, such as estimated cost of damage and name of responsible
person, are deleted. In the second round of data preparation, all the data
3.1. Description of the dataset with missing values are removed. In the third round, all data are checked
to ensure there is no error data and outliners. Finally, a total of 557
The data used in this research were provided by well-established projects and 23,574 injury events over 11 years was collected and used.
construction companies operating nation-wide and focused on all
types of construction contracting and service providers. As such, the 3.3. Data analysis methods and processes
data is highly representative. The dataset contains three parts of infor­
mation, namely basic information of time and the injured person, injury 3.3.1. Statistical analysis
detail description information and reason (causes) of injury or how the Statistical analysis is a science of collecting, exploring and presenting
injury happened. The basic structure of the dataset is as shown in Fig. 1. large amounts of data to uncover patterns and trends (Sprinthall and
From the basic information, we can know the age, gender and occupa­ Fisk, 1990). In this study, descriptive statistics analysis, probability
tion of the injured person, as well as the exact time of the injury. For distribution analysis and linear regression analysis are applied.
example, the injury numbered CDY-0010565 occurred at 11:40 pm on Descriptive statistics could describe the basic feature of the data and this
3rd April 2013, and the injured worker is a 25 year-old male concreter. study uses descriptive statistics to describe the injury proportions of
The injury information includes the nature of injury (e.g., superficial body parts. Probability distribution analysis is a method that allows us to
injury, muscle/tendon strain, contusion), bodily location (e.g., wrist, create a statistical distribution with a specific pattern to describe the
back, hand, leg and ankle), and detailed description of injury. In addi­ probability of injury. Moreover, from probability distribution, we can
tion, companies conducted in-depth investigations of the causes of every estimate the probability when (time) an injury event might be likely to
injury, and the content and report of investigation include event occur. Linear regression is applied to analyse variation over time in the
description and mechanism of injury (e.g., hitting moving objects, proportion of injuries due to different natures of injury. All data analysis
muscular stress while lifting, carrying or putting down objects, and were performed using R, a free software environment for statistical
hitting stationary objects). computing and graphics.
Data types included nominal data, text data and interval data.
Nominal data uses nominal scale to label variables without any quan­ 3.3.2. Data visualization
titative value, such as gender, occupation category, nature of injury, Data visualization refers to the techniques used to present data by
bodily location, and mechanism of injury. Text data usually consists of encoding it as visual objects in a graphic format. It is considered as a
sentences or paragraphs. In this study, event descriptions and detailed central part of exploratory data analysis (Myatt and Johnson, 2009).
descriptions of injury are text data. Interval data represent the numerical Compared with traditional descriptive statistics, data visualization can
value of variables, e.g., date, time and age. distil large datasets into graphics to allow for an easy understanding of
trends and complex relationships within the data. In this study, we apply

3
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

Fig. 2. Probability distribution of safety incidents by calendar month.

data visualization methods to visualize the results of the statistical Confidence(X→Y) = P(Y|X) (2)
analysis. First, probability distribution of safety incidents by month and
time of the day are presented, from which months and time of the day Lift is a measure of “interestingness” of a discovered rule. It is defined
with a higher frequency of safety incidents can be easily identified. as the correlation between the occurrence of X and Y(Eq. (3)). Specif­
Second, bodily locations of injury and corresponding injury proportions ically, a lift value equal to 1 means that the occurrence of X and Y are
are shown in the form of body structure diagrams. Third, age distribu­ independent of each other. A lift value greater than 1 indicates the
tions of the injured, as well as bodily locations of injury, are presented. occurrence of X and Y are dependent on each other, while a lift value
Forth, the relationship between injured body parts and mechanism of smaller than 1 means the occurrence of X and Y are substitute to each
injury, and the relationship between injured body parts and nature of other.
injury are presented in the form of scatterplots with smoothed densities.
P(X, Y) P(Y|X)
All data visualisation was performed using R, a free software environ­ Lift(A→B) = = (3)
P(X)P(Y) P(Y)
ment for statistical computing and graphics.
This research employs ARM to discover the rules between occupation
3.3.3. Association rule mining method category, nature of injury, bodily location, and mechanism of injury.
Association rule mining (ARM) is an unsupervised learning method Apriori algorithm is selected to perform ARM as it is for frequent item set
that is used to discover hidden relations and correlations in large data mining (Agarwal and Srikant, 1994). Rules with a support above 0.006
sets (Xiao and Fan, 2014). It was first introduced for discovering regu­ and confidence above 0.25 are first selected (Verma et al., 2014). After
larities between products in supermarket (Agrawal et al., 1993). that, rules are checked to ensure that whether they make sense. Un­
Currently, ARM has been applied in the field of sociology (Hastie et al., reasonable rules are eventually removed.
2009), chemistry (Singh and Singh, 2011), and building energy (Fan
et al., 2015). The rules are based on the frequency number of an item set 4. Results and discussion
in combination with other sets in a database (Verma et al., 2014). Brief
descriptions of the ARM method are provided in the following sections. 4.1. Time of injury
Let α be a large item set, an associate rule is defined as an implication
of the form X→Y, where X, Y ⊂ α, and X ∩ Y=∅. The set X is called the 4.1.1. Injury time of the year
antecedent and the set Y is called the consequent. There are two mea­ From the perspective of calendar months of the injury events,
surements of rule effectiveness, i.e. support and confidence (Verma February, October and November had higher percentages of injury
et al., 2014). Support reflects the usefulness of discovered rules, and it is (Fig. 2). The research data was in the southern hemisphere where
the joint probability of the antecedent and consequent (Eq. (1)). Confi­ February is hot summer and the long sunshine and long daily solar ra­
dence refers to certainty of discovered rules and it is an estimate of the diation make construction sites a very unpleasant working environment
conditional probability, as shown in Eq. (2). Only those rules that meet (Chiang et al., 2018). Further, the temperatures can reach to high 30 s
the threshold of support and confidence are considered to be meaningful degrees and becomes heatwaves. Xiang et al. (2014) found that a 1 ◦ C
(Fan et al., 2015). In this study, the threshold of support and confidence increase in temperature between 14.2 ◦ C and 37.7 ◦ C was associated
is set to 0.006 and 0.25, respectively. with a 0.2% increase in daily injury events. In addition, workers wear
less clothing during the summer months, making them more susceptible
Support(X→Y) = P(X, Y) (1)
to cuts, bumps and bruises. Compared with summer, injury incidents are

4
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

Fig. 3. Probability distribution of safety incidents by time of the day.

lower in winter (June, July and August). On one hand, the winter is adjust themselves to the pace of work. Otherwise, workers are likely to
relatively mild. This provides a more suitable environment for con­ be injured as they are not comfortable with this shift from vacation to
struction. Moreover, workers wear more clothing in winter, which also work.
protect workers from injury to some extent.
December and January are found to have a lower percentage of 4.1.2. Injury time of the day
injury, despite it was also summer time. The main reason may be that Construction workers usually start work at 7:00 or 7:30 in the
construction work is least intensive during the two-month period due to morning and finish at 15:30 or 16:00 in the afternoon Monday to Friday,
Christmas and New Year holidays. Similarly, construction work is less in and Saturday mornings. As shown in Fig. 3, the period between 7:00 am
April as most workers take the Easter holiday period off. It is interesting and 9:00 am witnessed a dramatic increase in the probability of injury
to find out that incidents increased rapidly after every holiday. This may event. The reason may be that workers’ minds were not with work at the
mean that workers need a period of adjustment after the holidays to beginning of the day. It is also found that most injury events occurred

Fig. 4. Proportion of injured body parts, organs and systems.

5
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

Fig. 5. Age distribution of the injured.

between 11:00 am and 12:00 noon. This was arguably due to the fact presented in Fig. 4. It can be seen that hand has the highest percentage of
that fatigue contributes to injury events. Fatigue level of construction injuries (27.75%), followed by back (11.59%), eye (10.39%) and foot
workers increases with the increase of the number of hours worked (6.72%). If extending the hand part to include arm, wrist and elbow, the
(Fang et al., 2015). If fatigue is not recovered, it could affect the physical percentage is 36.54%, which is over one-third of the injuries. Similarly
and cognitive functions of construction workers (Zhang et al., 2015). It is for foot, if extended to include leg, knee and hip, the percentage is as
found that there is a small decrease in the probability of injury events high as 19.95%. Based on the event descriptions, the main causes of
between 10:00 and 11:00. This is slightly different from findings in the hand injuries include “being hit by moving/falling objects”, “hitting
research of Chiang et al. (2018) and Huang and Hinze (2003) where they stationary objects”, “being trapped between stationary and moving ob­
stated that most accidents occurred between 10:00 am and 11:00 am. jects”, and “contact with electricity”. The very high rate of 27.75% hand
The main reason may be that construction workers usually have a short injury indicates the need for wearing glove at construction worksites.
break (morning tea) for drinking coffee during this period. These results Although some construction companies have introduced ‘wearing glove’
also highlight the need of taking breaks during the working day. as a mandatory requirement, most companies are still not making this
There was a significant decline from 12:00 to 13:00 in the afternoon. mandatory. This is a global problem that deserves more attention of
This aligns with the findings of Chiang et al. (2018) and Yi and Chan policy makers, construction managers, and researchers.
(2015), who reveal that the number of injury accidents decreases Back injuries have long been seen as one of the largest segments of
significantly during lunch break. After lunch break, however, there was worker injuries (Bernold and Guler, 1993; Lette et al., 2018). Most of
a small increase in the probability of injury events. The same phenom­ back injuries are sprains and strains linked to lifting, loading, carrying
enon also occurred after morning tea. This may be arguably due to the and pushing activities (Bernold and Guler, 1993). Being the second
fact that workers were not in a very good working state after morning highest rate of 11.59%, back injury can not be ignored by policy makers
tea and lunch break. On one hand, a short break could relieve physical and construction manager any longer. Something must be done to pro­
and mental fatigue of workers. On the other hand, working state may be tect this part of human body of the construction workers. Researchers
affected by a break. Therefore, it is crucial for construction workers to are also enraged to look into this problem.
get back their working state after break. Although eye protection, in the form of safety goggles, glasses and
Fig. 3 also shows that there are not many injuries occurred before face shields, have been available on construction sites (Hinze and Giang,
7:00 and after 17:00. This is mainly because no construction work is 2008), eye injuries continue to be disproportionately high. The main
normally allowed to be carried out during these time periods, unless it reasons of eye injuries included “exposure to other and unspecified
was essential to undertake a construction task during the outside normal environmental factors”, “single contact with chemical or substance”,
working hours (such as on-road/sea transportation of construction ma­ and “being hit by moving/falling objects”. As we know eye injury could
terials or equipment). Furthermore, if work is carried out outside the be very serious, and could be partial permanent loss. Wearing safety
normal 7:30–17:00 daily routing, the costs for worker’s wages would be googles/glasses should be made as mandatory requirement on con­
doubled which may significantly increase project costs. struction sites.
While safety boots on construction sites have been introduced long
4.2. Bodily locations, organs and systems of injuries ago, foot and ankle injuries still appear to be a major problem of total­
ling to 11.52%. There is a need to improve this situation, through policy,
4.2.1. Bodily locations and organs management and research. If also counting leg and knee, the total would
The proportion of injured body parts, organs and systems is be 24.19%. It should be pointed out that the upper parts of human body

6
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

Fig. 6. Relationships between age and bodily locations of injury.

such as shoulder, neck, face and head also suffer seriously from injuries, locomotor system, nervous system is more vulnerable. There are many
counting for over 12%; moreover, it should be pointed out that injuries causes of injury to nervous system, such as “mental stress”, “exposure to
occurred in most parts of human body as shown in Fig. 4. environmental heat”, “being trapped between stationary and moving
objects”, and “muscular stress”. Therefore, preventing neurological in­
4.2.2. Injured body systems juries is more difficult. There is a need to conduct further research on
Regarding the injured systems, as shown on the right hand-side this problem, by using multi-disciplinary methods that include neuro­
column of Fig. 4, locomotor system has the highest percentage of in­ science, psychology, medicine and construction safety management as
juries, at more than 88%, and nervous system accounted for over 11%. well as emergent information and communication technologies. The
These two systems together accounted for over 99% of injuries. Loco­ above findings are useful for workers training and also for on-site first
motor system is the main body system used by workers during con­ aid preparation including first aid officer training and corresponding
struction activities. It generally interacts or comes into direct contact medicine storing.
with the external environment. Construction workers usually have
heavy work, and thus their risk of injuries to the locomotor system is
4.3. Age distribution of the injured workers
much higher than that of other workers with less heavy work
(Schneider, 2001). Injuries to the locomotor system often appear dra­
The age of the workers who sustained injuries ranged from 18 to 70
matic, but most of them are not life-threatening (Alsheikhly and
(Fig. 5). The age distribution of the injured workers is broadly consistent
Alsheikhly, 2018). However, if they are not prevented and treated well,
with the findings of previous studies (Chiang et al., 2018; Hinze and
they are likely to have a lasting impact on the subsequent work and lives
Giang, 2008). Fig. 5 shows that workers under and around age of 20 had
of workers.
low injury rates. This may be due to they were being supervised by a
Injuries to nervous system also cannot be ignored. The nervous sys­
senior worker (particularly if they are an apprentice) hence more
tem is relevant to worker perception and control. Compared to
cautious. However, the trend increased rapidly from the mid-20 s and

Fig. 7. Relationship between injured body parts and mechanism (causes) of injury.

7
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

Fig. 8. Relationship between the injured body parts and nature of injury.

particularly between the ages of 25 and 30 are most likely to be injured 4.4. Causes and nature of injury relating to body parts
(Fig. 5). Lander et al. (2016) and Westaby and Lowe (2005) mentioned
that age can be considered as a strong risk predictor of injuries as the 4.4.1. Causes of injury relating to body parts
frequency of injuries is higher among young workers. Generally The relationship between injured body parts and causes of injury is
speaking, young workers are inexperienced and can easily overlook presented in Fig. 7. The horizontal coordinate indicates the injured body
potential safety risk factors. Mučenski et al. (2015) found that there is a part, and the vertical coordinate indicates the cause/mechanism of
trend of decrease in the occurrence of injuries along with the increase in injury. The colour indicates the number of overlapping points in the
working experience. Furthermore, younger workers have not experi­ graph: the darker colour of the area, the more numbers of overlapping
enced many safety incidents, and they are more likely to take risks and points of injured body part and causes of injury. For example, Area a
disobey safety rules and expose themselves to dangerous environments. contains hand injuries, and specifically, it shows that hand injuries are
In contrast, older workers showed more positive safety attitudes than strongly associated with hitting moving/stationary objectives, and being
younger ones (Siu et al., 2003). hit by moving/falling objectives. Although the number of overlapping
A significant negative correlation between age and the probability of points in Areas b, c and d is less than Area a, they still indicate possible
injury beyond the age of 30 is found, and there are mainly three possible links between injured body parts and mechanism of injury, and cannot
causes of this phenomenon. First, older workers have more experience be overlooked.
and are relatively with more safety conscious. Second, older workers are As shown in Fig. 7 the main causes of body injury include being hit by
aware that fewer job opportunities are available for them, and thus they objects, weight bearing, and hitting objects. More effort should be paid
are more committed at work and are willing to comply with safety rules to eliminate, reduce or mitigate these causes. However, that being said,
(Siu et al., 2003). Third, older workers might have more family or whenever an injury has happened, it could happen again in the future,
society-related responsibilities which remind them to avoid taking even if the probability of occurrence is relatively low. Therefore, suffi­
safety risks. It is noteworthy that, although older workers have lower cient attention should be given to every cause of injury. It is also found
rates of injury, there are more deaths among them due to the decline in that exposure to environmental cold or heat or a traumatic event is less
physical and psychosocial conditions (Amissah et al., 2019; Chiang likely to result in injury to body parts. On one hand, these sources of
et al., 2018; Schwatka et al., 2012). Based on the above knowledge danger were less likely to occur. On the other hand, workers were better
discovered and analysis, it is clear that there is a need to develop and protected against these sources of danger.
provide customised and personalised safety training to young con­
struction workers in the age group of 20–35, to improve their under­ 4.4.2. Nature of injury relating to body parts
standing, attitude and perception of safety risks. The relationships between injured body parts and nature of injury is
The relationships between age and bodily locations of injury are shown in Fig. 8. The horizontal coordinate indicates the injured body
presented in Fig. 6. Nine bodily locations are considered, i.e., ankle, arm, part, and the vertical coordinate indicates the nature of injury. Area a
back, eye, foot, hand, knee, leg and shoulder, as they have a higher indicates the most likely nature of injury to hand, back and eye.
percentage of injuries. In Fig. 6, each “violin” plot represents the prob­ Specially, hands are more likely to have superficial injury and laceration
ability distribution of age by a bodily location of injury. The white dot or open wound. The main reason includes but not limited to being hit by
represents the median age. For example, the white dot of ankle indicates object, contact with electricity, and hitting object. Compared with other
that the median age of workers with ankle injuries is 31. The thick black body parts, hands have the most diverse nature of injury. Hands are vital
bar in the centre shows the interquartile range. From Fig. 6, it can be to construction workers, especially for manual workers and craftsmen
found that the presented bodily locations of injury were mainly on (Garg et al., 2012). Due to the large amount of work requires the use of
young workers under 30 years. Meanwhile, older workers are more hands, they are more easily exposed to dangerous environment. Back
likely to have injuries to knee, shoulder, leg and back. This may be due injury is highly related to muscle/tendon strain. If muscle/tendon strain
to the fact that the functions of these body parts have declined as is not treated promptly, workers may experience recurring injuries or
workers getting older. Therefore, older workers should be adequately pain as well as weakness in the muscle at work. Eye is more likely to
protected in these body parts. suffer from “foreign body on external eye”, which is usually caused by
being hit by objects or exposure.
Different natures and types of injury may have varying degrees of
impact on workers. For example, workers with superficial injuries may

8
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

Table 1 fractures, should be taken seriously. It is also suggested that the design of
The top 20 results of association rule mining. PPE should take into account the different nature of injury to different
No. Rules Support Confidence Lift body parts.
1 Hand, Contact with electricity => 0.80% 87.57% 78.91
Electrocution shock from electric current 4.4.3. Results of association rule mining (ARM) analysis
2 Contact with electricity => 1.08% 85.37% 76.92 In total 128 rules were derived from the datasets based on associa­
Electrocution shock from electric current tion rule mining methods (for the full list, please see Appendix A). These
3 Contact with hot objects => Burn 2.70% 93.23% 23.92 128 rules describe in detail the possible injuries to different body parts
4 Tradesmen, Eye => Exposure to other 2.95% 60.21% 12.09
and unspecified environmental
as well as the corresponding causes, providing a basis for workers and
conditions management to develop targeted protective measures. The top 20 rules
5 Tradesmen, Eye, Exposure to other and 2.85% 96.50% 10.42 are summarised in Table 1, which shows not only the relationship be­
unspecified environmental conditions tween the injury and body parts, but also the causes/mechanism of the
=> Foreign body on external eye,
injuries. Most of the rules may seem to be common sense and can be
6 Tradesmen, Eye => Foreign body on 4.39% 89.47% 9.66
external eye obtained from work experience and domain knowledge. However, they
7 Exposure to other and unspecified 4.35% 87.25% 9.42 reflect that these rules are common on construction sites and cannot be
environmental conditions => Foreign ignored. For instance, Rule 1 in Table 2 states that, if workers’ hand is in
body on external eye in ear or nose or in contact with electricity, they could have an electrocution shock from
respiratory digestive or reproductive
tract
electric current. This can be easily interpreted, as construction sites are
8 Eye => Foreign body on external eye in 8.40% 85.91% 9.28 prone to electrical wiring leaks and workers mainly use their hands to
ear or nose or in respiratory digestive or operate equipment at construction sites. As another example, Rule 8
reproductive tract describes that, workers’ eyes are more likely to sustain “foreign body on
9 Eye => Exposure to other and 4.48% 45.80% 9.20
external eye”. This can also be easily understood, as there is dust on the
unspecified environmental conditions
10 Eye, Single contact with chemical or 1.12% 72.58% 7.84 construction site and it could easily get into the eyes of workers who are
substance => Foreign body on external not wearing protective glasses. For example, Rule 19 indicates that face
eye may be hit by moving objects. Accordingly, workers are suggested to
11 Ankle => Stepping kneeling or sitting on 2.46% 47.41% 7.44 wear face protection on construction sites where there are moving
objects
objects.
12 Single contact with chemical or 0.77% 26.28% 6.74
substance => Burn As shown in Table 1, most injuries happened with hand, back, eye,
13 Tradesmen, Knee => Stepping kneeling 0.73% 29.68% 4.66 face and foot (include ankle, knee) due to different causes, which
or sitting on objects resulted in similar conclusion as discussed in previous sections.
14 Back => Muscular stress while lifting 4.28% 34.93% 4.51
carrying or putting down objects
15 Tradesmen, Back, Muscular stress while 1.34% 99.24% 4.37 4.5. A decade long changing trends of incidents
handling objects other than lifting
carrying or putting down => Muscle/ Fig. 9 shows the changing trends of incidents over a period of 11
tendon strain (non traumatic) years, which is indicated by the correlations expressed by the linear
16 Tradesmen, Muscular stress while lifting 2.66% 93.45% 4.11
carrying or putting down objects =>
regression line between different types of incidents and time. Y-axis
Muscle/tendon strain (non traumatic) indicates the number of a particular type of injury as a proportion of the
17 Tradesmen, Back => Muscle/tendon 4.25% 90.85% 4.00 total number of injuries.
strain (non traumatic) The matching linear coefficients and standard errors (SE), t-values
18 Hand, Hitting stationary objects => 4.34% 43.19% 2.89
and p-values are shown in Table 2. It is found that the p-values of con­
Laceration or open wound not involving
traumatic amputation tact, fall, and muscular stress are less than a significant level of 0.05,
19 Face => Being hit by moving objects 1.49% 46.45% 2.85 which indicates that there are significant correlations between time and
20 Hand => Laceration or open wound not 8.84% 32.92% 2.20 injuries due to contact, fall and muscular stress. Specifically, the 11
involving traumatic amputation years period witnessed a decrease of proportion of injuries due to con­
tact. This indicates that construction companies have been effective in
preventing worker injuries due to contact with or exposure to other
Table 2 objects during this decade. This could be due to enforcement of use of
Single variable regression for proportion of causes of injury and time. personal protective equipment (PPE). Indeed, most injuries due to
Reason Coefficient SE t value p value contact (e.g., contact with biological items, contact with poisonous parts
Being hit − 0.3713 0.3054 − 1.215 0.255 of plant or marine life, and contact with hot objects) can be effectively
Contact − 0.5339 0.1872 − 2.852 0.019* prevented as long as workers wear protective equipment (Hinze and
Exposure 0.1365 0.2381 0.573 0.5806 Giang, 2008). The percentage of injuries due to muscular stress and fall
Fall 0.2711 0.1048 2.588 0.0293* is increasing over time, both of which should be taken into consideration
Hitting − 0.3385 0.2546 − 1.329 0.217
seriously by the management and workers. Falling from height has al­
Injured by wildlife 0.0520 0.0682 0.762 0.4658
Muscular stress 0.3920 0.1263 3.103 0.0127* ways been a major safety problem on construction sites. It is unfortunate
Repetitive movement, low muscle 0.0609 0.0504 1.208 0.2577 that it still remains as a major safety problem. More research should be
loading conducted to find effective solutions. The causes of muscular stress
Stepping, kneeling or sitting on 0.1126 0.1050 1.072 0.3115
should also be investigated, and the same goes to repeat movement,
objects
Vehicle accident − 0.1773 0.1674 − 1.06 0.3169
which also has shown an increasing trend over the period.
Apart from the above statistical analysis results, the general trends
Note: * indicates p-value < 0.05
are that ‘hitting or being hit by objects’ remained high, despite some
reduction over the period. Several other types of incidents were also
be able to return to work with a simple bandage, while workers with remained high in the first five years then reduced in the second five
joint dislocation may lose workdays. Therefore, some nature of injury years, possibly due to some preventive measures. Such incidents include
with potentially serious consequences, such as burn, dislocation and ‘stepping, kneeing and sitting on objects’, ‘vehicle accident’ and ‘expo­
sure’. Some other types of incidents, such as ‘repetitive movement’ and

9
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

Fig. 9. Correlations by linear regression between natures of injury and time (years).

‘low muscle loading’ had irregular trends over the 11 years period. after an incident or injury has occurred. Strategies 4 and 5 are used for
Attention should be paid to “injured by wildlife”, particularly for con­ upgrading the level of safety protection. The five strategies are inte­
struction of horizontal/line project such as highway, road and rail. grated as shown in Fig. 10, while each of these strategies is discussed in
detail in the following sections.
5. Strategies for improving safety management
5.1. Providing personalised safety training (Strategy 1)
Based on the causes, mechanism and knowledge discovered and
presented in Section 4, several strategies are proposed to reduce or Safety training is considered as an effective safety management
eliminate the causes and risks and improve safety: (1) providing diver­ method as it could accelerate workers’ safety knowledge and skill
sified personalised safety training to all levels of workforce and man­ development. However, the existing forms of safety training was criti­
agement; (2) conducting a short (5–10 min) safety talk at strategic points cised as being boring and lack intuition and interaction (Cheng et al.,
to raise safety awareness; (3) providing tailored medical training for first 2004). There is a need to change the method of safety training (Li et al
aid officers and workforce; (4) making wearing personal protective 2018). Therefore, it is suggested to provide diversified and personalised
equipment (PPE) mandatory for site workers; (5) enhancing real-time safety training. For example, a safety training session could be devel­
site monitoring and safety warning by applying digital information, oped in according to the findings of this research. Nowadays, virtual
sensing and communication technologies. Strategies 1 and 2 are used for reality (VR) technology have been developed to reveal the construction
improving workers’ safety skills and awareness according to the time of process and site environment for improving safety consciousness of
injury and causes of injury. Strategy 3 is used for emergency response workers (Guo et al., 2017; Li et al., 2018). The VR technology can

10
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

5.3. Implementing tailored first-aid training and medical preparation


(Strategy 3)

According to the results of this research, construction workers may


suffer injuries from various parts of their bodies, organs and systems.
First-aid training should focus on these areas, especially hand, back, leg,
and foot. Furthermore, workers are suggested to complete first aid
training to prepare for onsite emergencies treatments. Experienced
workers and managers should also expand their knowledge in basic first
aid. More importantly, medicine storing and medical preparation should
pay particular attention to these specific areas and store medicine
accordingly. It is highly recommended that medication should be pre­
pared according to the nature of injury, especially for muscle/tendon
strain, laceration and open wound, superficial injury, and foreign body
on external eye. In this way, construction companies can allocate limited
healthcare resources to where they are most needed.

5.4. Making wearing personal protective equipment mandatory on site


(Strategy 4)

Although wearing safety personal protective equipment (PPE)


cannot guarantee that workers are completely safe from injury, they can
reduce the extent to which a worker is injured. This study found that
hand, foot, back and eye are the most vulnerable bodily parts. Most
Fig. 10. Integrated framework for improving safety in construction. injuries to these bodily parts can be mitigated by wearing PPE. For
example, gloves, booths, safety glasses, safety belts and face shields
provide an opportunity for workers to practice how to do a dangerous could protect workers from “exposure to other and unspecified envi­
task in a virtual environment, without any real risk (Lu and Davis, ronmental conditions”, “single contact with chemical or substance”. In
2016). fact the requirement to wear PPE in construction sites are not new and
Construction companies could apply the knowledge discovered in has been implemented for a long time. However, there is lack of con­
this current research into a VR scenario as safety training resources. For sistency, particularly on hand and back protection. Safety PPE can be
instance, safety manager could design all scenarios related to hand in­ regarded as workers’ last line of defence to prevent and mitigate in­
juries due to “being hit by moving objects” in a virtual environment. In juries. The manufacture of more injury-specific PPE should be encour­
this way of training, workers could learn to protect their hands from aged. Site workers must be required as mandatory to wear PPE. For aged
being hit by moving objects. This would allow workers to reinforce workers, more attention should be paid on knee, shoulder, leg and back
safety awareness in a virtual learning experience. as they are more likely to have injuries. PPE for these bodily parts should
be designed to reduce the likelihood of injury to aged workers.
5.2. Conducting short safety meetings at strategic points (Strategy 2)
5.5. Enhancing real-time on-site monitoring by applying emerging digital
One of the most important findings is that there are four peak times technologies (Strategy 5)
of incidents occurring by calendar months and three peak times during a
working day. With regard to month, incidents are more likely to happen There are diverse types of digital information and communication
before and after holidays (e.g., Christmas Holiday and Easter Holiday). technologies (ICTs) that can be applied in safety management at present,
The main reason may be that workers are not in a very good working such as Radio Frequency Identification (RFID), bar codes, Global Posi­
mood and state, hence not work-ready before and after holidays. tioning System (GPS), 5G, Internet of Things, sensors, ultra-wide band,
Regarding time of the day, incidents are more likely to occur before and camera, 3D laser scanner and BIM (Skibniewski, 2015; Zhang et al.,
after morning-tea and lunch breaks. Specifically, before a break, workers 2017). These technologies could collect real-time construction site
may be in a hurry to have their break. After a break, workers may feel safety information and transmit the information to the right person at
tired and not ready to re-start their work. the right time (Zhou et al., 2013). These ICTs can improve safety man­
A possible solution to the above-mentioned problems is to have a agement in three ways. First, they can be used to alert workers to the
short 5–10 min safety meeting at the above-mentioned strategic points four peak times of incidents by calendar month and three peak times by
of time. There are three main reasons on limiting meetings to 5–10 min. time of a working day identified in this study. For example, sensors can
First, the attention span of people is no more than 15 min (Johnstone be attached to workers’ helmets, and it could alert workers at these
and Percival, 1976). If the safety meeting is too long, workers’ attention strategic time points. Second, they can help workers identify sources of
is likely to be distracted (McLeish, 1968). Second, construction projects danger on construction sites. Specifically, sensors, ultra-wide band or
are strictly controlled in terms of schedule (Xu et al., 2018b). Prolonged RFID can be installed near the locations of the sources of danger, and
safety meetings often affect the project schedule. Third, this meetings sensors worn by workers will alert if they are near a source of danger.
serve to recall and remind the workers of the importance of safety and For instance, when a worker approaches a moving object, the sensor
enhance their safety awareness, rather than technical contents-focused. could alert the worker to the potential injury to the worker from the
During the meeting, workers’ safety awareness can be enhanced and moving object. Third, sensors, cameras, RFID can be also used to check
they can also have some time to adjust to the normal pace of work. whether workers are wearing PPE as required. In addition, the data
Meanwhile, workers can get a proper break in the meeting. Especially collected during this process will form good dataset and can be further
the aged workers who could have a rest to recover from the strain of site analysed to discover new safety knowledge and mechanisms.
work (Chiang et al., 2018). This 5/10-minute meeting could also serve as
a transition period for the workers to emotionally get ready to go back to
work on their construction sites.

11
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

6. Research implications and contributions Table A1


Full set of results of Association Rule Mining.
The implementation of this research is manifested in two aspects. No. Rules Support Confidence Lift
Firstly, major safety risks (i.e., causes of incidents and injuries) are
1 {Hand, Contact with electricity} => 0.80% 87.57% 78.91
unfolded from this research. Safety incidents and injuries that have {Electrocution shock from electric
occurred in the past are likely to occur again in the future, if nothing is current}
done to manage the safety risks. The results of this study show not only 2 {Contact with electricity} => 1.08% 85.37% 76.92
the safety risks with a high probability of occurrence, but also the risks {Electrocution shock from electric
current}
with a lower probability of occurrence that may be overlooked. Sec­ 3 {Hand, Contact with hot objects} => 1.24% 96.40% 24.74
ondly, this study provides a valuable reference for applying combination {Burn}
of new data analysis methods to the field of safety science and man­ 4 {Tradesmen, Hand, Contact with hot 0.61% 93.70% 24.04
agement. A combination of statistical analysis, data visualisation and objects} => {Burn}
5 {Contact with hot objects} => {Burn} 2.70% 93.23% 23.92
association rule analysis methods was applied. The results obtained from
6 {Tradesmen, Contact with hot objects} 1.39% 90.27% 23.16
different analytical methods can be verified and complemented by each => {Burn}
other, which enhances the systematic and comprehensive nature of the 7 {Tradesmen, Eye} => {Exposure to 2.95% 60.21% 12.09
research. other and unspecified environmental
This study contributes to academic outcomes by provides a “pano­ factors}
8 {Tradesmen, Eye, Exposure to other 2.85% 96.50% 10.42
rama” yet detail and in-deepth understanding of construction injury and unspecified environmental
based on mining and analysing a large data set with consideration of factors} => {Foreign body on external
time of the accidents, bodily location of injury, age of the injured, and eye in ear or nose or in respiratory
characteristics, nature, cause and mechanism of injury. A total of 19 digestive or reproductive tract}
9 {Eye, Exposure to other and 4.24% 94.59% 10.21
injured body parts, 11 injured organs and 8 injured systems, 28 causes of
unspecified environmental factors} =>
injuries and 16 nature of injury were analysed, which is more detailed {Foreign body on external eye in ear or
than the previous studies. Researchers could benefit from gaining a nose or in respiratory digestive or
better understanding of the characteristics of injury and the formation reproductive tract}
mechanism of injury. 10 {Tradesmen, Exposure to other and 2.92% 94.32% 10.19
unspecified environmental factors} =>
From the perspective of practice, the knowledge discovered and
{Foreign body on external eye in ear or
strategies proposed in this study could help practitioners better target nose or in respiratory digestive or
their efforts to mitigate workplace injuries risks and prevent incidents reproductive tract}
from happening. Based on the results of this study, project managers and 11 {Tradesmen, Eye, Being hit by moving 0.84% 91.01% 9.83
objects} => {Foreign body on external
workers could know which time periods are at higher risk of injury and
eye in ear or nose or in respiratory
which body parts are more vulnerable to injury, so that more effective digestive or reproductive tract}
countermeasures can be applied. Furthermore, safety skills, safety 12 {Tradesmen, Eye} => {Foreign body 4.39% 89.47% 9.66
awareness, safety protection, and emergency response are considered in on external eye in ear or nose or in
the proposed safety improvement strategies, providing all-round safety respiratory digestive or reproductive
tract}
protection for workers.
13 {Eye, Being hit by moving objects} => 2.02% 87.87% 9.49
{Foreign body on external eye in ear or
7. Summary, future research directions and conclusion nose or in respiratory digestive or
reproductive tract}
14 {Exposure to other and unspecified 4.35% 87.25% 9.42
To gain a systematic and comprehensive understanding of injury
environmental factors} => {Foreign
causes, nature and formation mechanism and insightful safety knowl­ body on external eye in ear or nose or
edge in construction, this research has used a large set of real data and in respiratory digestive or reproductive
investigated the characteristics, trends, causes and mechanisms of con­ tract}“
struction injury and accidents, with different age groups, based on sta­ 15 (Zhang et al.) => {Foreign body on 8.40% 85.91% 9.28
external eye in ear or nose or in
tistical analysis, data visualisation and association rule mining of the respiratory digestive or reproductive
injury events, and discovered the underlying knowledge. From the tract}
perspective of time of injury, it is found that there are three peak months 16 (Zhang et al.) => {Exposure to other 4.48% 45.80% 9.20
of injury /incident rates during a year and three peak times during a and unspecified environmental
factors}“
working day. Regarding bodily parts, hand, back and eye are most likely
17 {Labourers, Eye} => {Foreign body on 0.91% 85.10% 9.19
to suffer injuries, followed by foot and human locomotor system which external eye in ear or nose or in
together have the highest percentage of injuries, of over 88%. Specif­ respiratory digestive or reproductive
ically for causes of injury, hands mainly suffer from “hitting moving and tract}
stationary objectives” and “being hit by moving/falling objectives”, and 18 {Eye, Single contact with chemical or 1.12% 72.58% 7.84
substance} => {Foreign body on
hands are more likely to have superficial injury and laceration or open external eye in ear or nose or in
wound. Back injury is highly related to “muscle/tendon strain due to respiratory digestive or reproductive
muscular stress”. Eye is more likely to suffer from “foreign body on tract}
external eye due to exposure”. Regarding workers’ age, workers who are 19 {Tradesmen, Ankle} => {Stepping 0.81% 49.06% 7.70
kneeling or sitting on objects}
between the ages of 25 and 30 have a higher percentage of injury
20 {Ankle} => {Stepping kneeling or 2.46% 47.41% 7.44
overall, while the older workers are more likely to have injuries to knee, sitting on objects}
shoulder, leg and back. Based on association rule mining (ARM) method, 21 {Single contact with chemical or 0.77% 26.28% 6.74
a total of 128 rules was derived from the datasets, which depicts the substance} => {Burn}
relationship between occupation category, nature of injury, bodily 22 {Tradesmenb1 = Knee} => {c1 = 0.73% 29.68% 4.66
Stepping kneeling or sitting on objects}
location, and causes of injury. 23 4.28% 34.93% 4.51
On the basis of the discovered knowledge, an integrated framework
(continued on next page)
that includes five strategies are proposed to improve construction safety
performance. They are (1) providing diversified personalised safety

12
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

Table A1 (continued ) Table A1 (continued )


No. Rules Support Confidence Lift No. Rules Support Confidence Lift

{Back} => {Muscular stress while {Tradesmen, Stepping kneeling or


lifting carrying or putting down sitting on objects} => {Muscle/tendon
objects} strain (non traumatic)}
24 {Tradesmen, Back, Muscular stress 0.66% 100.00% 4.40 46 {Hand, Hitting stationary objects} => 4.34% 43.19% 2.89
with no objects being handled} => {Laceration or open wound not
{Muscle/tendon strain (non involving traumatic amputation}
traumatic)} 47 {Tradesmen, Hand, Hitting stationary 2.15% 42.99% 2.88
25 {Single contact with chemical or 1.19% 40.56% 4.38 objects} => {Laceration or open
substance} => {Foreign body on wound not involving traumatic
external eye in ear or nose or in amputation}
respiratory digestive or reproductive 48 {Face} => {Being hit by moving 1.49% 46.45% 2.85
tract} objects}
26 {Tradesmen, Back, Muscular stress 1.34% 99.24% 4.37 49 {Shoulder, Muscular stress while 0.74% 64.13% 2.82
while handling objects other than lifting carrying or putting down
lifting carrying or putting down} => objects} => {Muscle/tendon strain
{Muscle/tendon strain (non (non traumatic)}
traumatic)} 50 {Tradesmen, Face} => {Being hit by 0.70% 45.03% 2.76
27 {Tradesmen, Back, Muscular stress 1.46% 98.61% 4.34 moving objects}
while lifting carrying or putting down 51 {Being hit by falling objects} => 1.08% 25.61% 2.76
objects} => {Muscle/tendon strain {Contusion bruising and superficial
(non traumatic)} crushing}“
28 {Tradesmen, Muscular stress with no 2.13% 95.38% 4.20 52 {Muscular stress with no objects being 3.44% 62.01% 2.73
objects being handled} => {Muscle/ handled} => {Muscle/tendon strain
tendon strain (non traumatic)} (non traumatic)}
29 {Labourers, Muscular stress while 1.13% 95.20% 4.19 53 {Back, Muscular stress with no objects 1.16% 61.20% 2.69
handling objects other than lifting being handled} => {Muscle/tendon
carrying or putting down} => strain (non traumatic)}
{Muscle/tendon strain (non 54 {Labourers, Hitting stationary objects} 0.97% 38.68% 2.59
traumatic)} => {Laceration or open wound not
30 {Tradesmen, Ankle, Stepping kneeling 0.76% 94.87% 4.17 involving traumatic amputation}
or sitting on objects} => {Muscle/ 55 {Back} => {Muscle/tendon strain 7.22% 58.88% 2.59
tendon strain (non traumatic)} (non traumatic)}
31 {Tradesmen, Muscular stress while 2.65% 93.45% 4.11 56 {Tradesmen, Knee} => {Muscle/ 1.43% 58.53% 2.57
lifting carrying or putting down tendon strain (non traumatic)}
objects} => {Muscle/tendon strain 57 {Face, Being hit by moving objects} 0.65% 43.40% 2.55
(non traumatic)} => {Superficial injury}
32 {Tradesmen, Muscular stress while 3.00% 92.96% 4.09 58 {Repetitive movement low muscle 1.36% 57.80% 2.54
handling objects other than lifting loading} => {Muscle/tendon strain
carrying or putting down} => (non traumatic)}“
{Muscle/tendon strain (non 59 {Back, Muscular stress while lifting 2.47% 57.71% 2.54
traumatic)} carrying or putting down objects} =>
33 {Tradesmen, Back} => {Muscular 1.48% 31.64% 4.09 {Muscle/tendon strain (non
stress while lifting carrying or putting traumatic)}
down objects} 60 {Muscular stress while lifting carrying 4.46% 57.60% 2.53
34 {Tradesmen, Back} => {Muscular 1.35% 28.89% 4.08 or putting down objects} => {Muscle/
stress while handling objects other tendon strain (non traumatic)}“
than lifting carrying or putting down} 61 {Leg, Hitting stationary objects} => 0.65% 37.54% 2.51
35 {Tradesmen, Repetitive movement low 0.90% 91.58% 4.03 {Laceration or open wound not
muscle loading} => {Muscle/tendon involving traumatic amputation}
strain (non traumatic)} 62 {Tradesmen, Face} => {Superficial 0.67% 42.72% 2.51
36 {Tradesmen, Back} => {Muscle/ 4.25% 90.85% 4.00 injury}
tendon strain (non traumatic)} 63 {Hand, Being hit by moving objects} 2.61% 37.44% 2.51
37 {Labourers, Muscular stress while 0.83% 89.94% 3.96 => {Laceration or open wound not
lifting carrying or putting down involving traumatic amputation}
objects} => {Muscle/tendon strain 64 {Tradesmen, Hand, Being hit by 1.11% 36.92% 2.47
(non traumatic)} moving objects} => {Laceration or
38 {Labourers, Back} => {Muscle/tendon 1.48% 89.69% 3.95 open wound not involving traumatic
strain (non traumatic)} amputation}
39 {Tradesmen, Ankle} => {Muscle/ 1.43% 87.42% 3.85 65 {Ankle, Stepping kneeling or sitting on 1.38% 56.09% 2.47
tendon strain (non traumatic)} objects} => {Muscle/tendon strain
40 {Tradesmen, Neck} => {Muscle/ 1.08% 78.36% 3.45 (non traumatic)}
tendon strain (non traumatic)} 66 {Tradesmen, Head} => {Hitting 0.70% 48.57% 2.46
41 {Shoulder} => {Muscular stress while 1.15% 26.08% 3.37 stationary objects}
lifting or putting down objects}“ 67 {Leg, Hitting stationary objects} => 0.72% 41.74% 2.45
42 {Tradesmen, Shoulder} => {Muscle/ 1.46% 75.20% 3.31 {Superficial injury}
tendon strain (non traumatic)} 68 {Labourers, Hand} => {Laceration or 1.16% 35.83% 2.40
43 {Back, Muscular stress while handling 2.26% 74.11% 3.26 open wound not involving traumatic
objects other than lifting carrying or amputation}
putting down} => {Muscle/tendon 69 {Tradesmen, Hitting stationary 3.51% 35.47% 2.38
strain (non traumatic)} objects} => {Laceration or open
44 {Shoulder, Muscular stress while 0.77% 70.42% 3.10 wound not involving traumatic
handling objects other than lifting amputation}
carrying or putting down} => 70 {Knee, Stepping kneeling or sitting on 0.77% 53.79% 2.37
{Muscle/tendon strain (non objects} => {Muscle/tendon strain
traumatic)} (non traumatic)}
45 1.93% 68.62% 3.02 71 2.37% 53.68% 2.36
(continued on next page)

13
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

Table A1 (continued ) Table A1 (continued )


No. Rules Support Confidence Lift No. Rules Support Confidence Lift

{Shoulder} => {Muscle/tendon strain {Knee} => {Muscle/tendon strain


(non traumatic)} (non traumatic)}“
72 {Hitting stationary objects} => 6.93% 35.04% 2.35 104 {Hand} => {Superficial injury} 7.29% 27.17% 1.60
{Laceration or open wound not 105 {Hand} => {Being hit by moving 6.96% 25.94% 1.59
involving traumatic amputation} objects}
73 {Ankle} => {Muscle/tendon strain 2.74% 52.89% 2.33 106 {Tradesmen, Hand} => {Being hit by 3.02% 25.79% 1.58
(non traumatic)} moving objects}
74 {Face} => {Superficial injury} 1.23% 38.55% 2.27 107 {Labourers, Hand} => {Superficial 0.87% 26.75% 1.57
75 {Tradesmen, Hand} => {Laceration or 3.88% 33.11% 2.22 injury}
open wound not involving traumatic 108 {Head} => {Being hit by moving 0.77% 25.38% 1.56
amputation} objects}
76 {Hand} => {d1 = Laceration or open 8.84% 32.92% 2.20 109 {Labourer, Hand} => {Hitting 0.75% 30.23% 1.53
wound not involving traumatic stationary objects}
amputation} 110 {Tradesmen, Leg} => {Superficial 0.89% 25.82% 1.52
77 {Neck} => {Muscle/tendon strain 1.51% 49.91% 2.20 injury}
(non traumatic)}“ 111 {Tradesmen, Limb} => {Hitting 0.77% 29.92% 1.51
78 {Tradesmen, Hand, Hitting stationary 1.86% 37.22% 2.19 stationary objects}
objects} => {Superficial injury} 112 {Being hit by moving objects} => 4.19% 25.68% 1.51
79 {Ankle, Falls on the same level} => 0.61% 49.58% 2.18 {Superficial injury}
{Muscle/tendon strain (non 113 {Labourers} => {Muscle/tendon 4.34% 34.26% 1.51
traumatic)} strain (non traumatic)}
80 {Tradesmen, Hand} => {Hitting 5.01% 42.77% 2.16 114 {Labourers, Being hit by moving 0.68% 25.54% 1.50
stationary objects} objects} => {Superficial injury}
81 {Labourer, Hand} => {Laceration or 0.79% 31.68% 2.12 115 {Tradesmen, Leg} => {Hitting 1.02% 29.55% 1.49
open wound not involving traumatic stationary objects}
amputation} 116 {Leg} => {Superficial injury}“ 1.77% 25.07% 1.47
82 {Face} => {Laceration or open wound 1.01% 31.61% 2.12 117 {Tradesmen} => {Muscle/tendon 13.82% 32.37% 1.42
not involving traumatic amputation} strain (non traumatic)}
83 {Tradesmen, Hitting stationary 3.55% 35.84% 2.11 118 {Wrist} => {Muscle/tendon strain 0.92% 31.50% 1.39
objects} => {Superficial injury} (non traumatic)}
84 {Hand, Hitting stationary objects} => 3.58% 35.63% 2.10 119 {Limb} => {Hitting stationary 1.11% 27.08% 1.37
{Superficial injury} objects}
85 {Tradesmen, Falls on the same level} 1.24% 47.52% 2.09 120 {Trunk} => {Muscle/tendon strain 0.81% 30.97% 1.36
=> {Muscle/tendon strain (non (non traumatic)}
traumatic)} 121 {Falls on the same level} => {Muscle/ 2.36% 30.61% 1.35
86 {Stepping kneeling or sitting on 3.00% 47.08% 2.07 tendon strain (non traumatic)}“
objects} => {Muscle/tendon strain 122 {Tradesmen, Limb} => {Muscle/ 0.78% 30.52% 1.34
(non traumatic)} tendon strain (non traumatic)}
87 {Head} => {Laceration or open wound 0.94% 30.80% 2.06 123 {Neck} => {Hitting stationary 0.79% 26.24% 1.33
not involving traumatic amputation} objects}
88 {Labourers, Hand} => {Hitting 1.32% 40.61% 2.05 124 {Limb} => {Muscle/tendon strain 1.09% 26.57% 1.17
stationary objects} (non traumatic)}
89 {Head} => {Hitting stationary 1.22% 40.10% 2.03 125 {Labourer} => {Hand}“ 2.49% 31.24% 1.16
objects} 126 {Tradesmen, Leg} => {Muscle/tendon 0.90% 26.12% 1.15
90 {Hitting stationary objects} => 6.71% 33.92% 1.99 strain (non traumatic)}
{Superficial injury} 127 {Tradesmen} => {Hand} 11.71% 27.43% 1.02
91 {Tradesmen, Hand, Being hit by 1.01% 33.50% 1.97 128 {Labourers} => {Hand} 3.24% 25.58% 0.95
moving objects} => {Superficial
injury}
92 {Labourers, Hand} => {Being hit by 1.02% 31.53% 1.93 training to all levels of workforce and management; (2) conducting a
moving objects} short (5–10 min) safety talk at strategic points to raise safety awareness;
93 {Labourers, Being hit by moving 0.76% 28.85% 1.93
objects} => {Laceration or open
(3) providing tailored medical training for first-aid officers and work­
wound not involving traumatic force; (4) making wearing personal protective equipment (PPE)
amputation} mandatory for site workers; (5) enhancing real-time site monitoring and
94 {Labourers, Hitting stationary objects} 0.81% 32.30% 1.90 safety warning by applying emerging digital information and commu­
=> {Superficial injury}
nication technologies.
95 {Hand} => {Hitting stationary 10.04% 37.40% 1.89
objects} Although this study has applied statistical analysis, data visualisation
96 {Being hit by moving objects} => 4.57% 28.06% 1.88 and association rule mining methods to discover injury-related knowl­
{Laceration or open wound not edge, there are other data mining methods, for example, cluster detec­
involving traumatic amputation} tion, decision tree and classification could also be used to discover more
97 {Tradesmen, Being hit by moving 1.89% 28.05% 1.88
objects} => {Laceration or open
knowledge. Future research could also focus on how to remove the root
wound not involving traumatic causes of injury that have been identified in this research.
amputation} In conclusion, by using a large set of longitudinal 11-years real data,
98 {Professional} => {Muscle/tendon 1.48% 41.75% 1.84 the findings of this research has confirmed that, despite some small
strain (non traumatic)}
changes over a decade, most injuries happened to hands, foot, back, eye
99 {Head} => {Superficial injury} 0.92% 30.29% 1.78
100 {Tradesmen, Being hit by moving 2.01% 29.89% 1.76 and face, and therefore the need to wear suitable PPE. Furthermore, this
objects} => {Superficial injury} research uncovered, in details, the fact that external environment and
101 {Tradesmen, Hand} => {Superficial 3.48% 29.72% 1.75 conditions have a major effect on workers safety, such as moving ob­
injury} jects, dust, electricity, temperature, which means site condition must be
102 {Hand, Being hit by moving objects} 2.03% 29.13% 1.71
=> {Superficial injury}
monitored and improved. In addition, workers’ mindsets, particularly
103 2.29% 36.51% 1.61 workers in their age of 20 s-30 s are more likely to be involved in ac­
cidents and being injured, tailored safety training is necessary for them.

14
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

Finally, it is necessary to look beyond the surface impact of incidents, Gebremeskel, T.G., Yimer, T., 2019. Prevalence of occupational injury and associated
factors among building construction workers in Dessie town, Northeast Ethiopia;
accidents and injuries, and into the long-term effect on human body,
2018. BMC Research Notes 12, 481.
organ and systems, as some injuries may lead to long-lasting impact on Guo, B.H.W., Zou, Y., Fang, Y., Goh, Y.M., Zou, P.X.W., 2021. Computer vision
them. technologies for safety science and management in construction: A critical review
and future research directions. Saf. Sci. 135, 105130. https://doi.org/10.1016/j.
ssci.2020.105130.
Declaration of Competing Interest Guo, H., Yu, Y., Skitmore, M., 2017. Visualization technology-based construction safety
management: A review. Autom. Constr. 73, 135–144.
The authors declare that they have no known competing financial Hallowell, M., Esmaeili, B., Chinowsky, P., 2011. Safety risk interactions among highway
construction work tasks. Construct. Manage. Econ. 29 (4), 417–429.
interests or personal relationships that could have appeared to influence Hastie, T., Tibshirani, R., Friedman, J., 2009. The elements of statistical learning: data
the work reported in this paper. mining, inference, and prediction. Springer Science & Business Media.
Hinze, J., Giang, G., 2008. Factors associated with construction worker eye injuries. Saf.
Sci. 46 (4), 634–645.
Acknowledgement Huang, X., Hinze, J., 2003. Analysis of construction worker fall accidents. J. Construct.
Eng. Manage. 129 (3), 262–271.
The authors gratefully acknowledge the support of this research by Isaac, S., Edrei, T., 2016. A statistical model for dynamic safety risk control on
construction sites. Autom. Constr. 63, 66–78.
the Fundamental Research Funds for the Central University (Grant No. Jin, R., Zou, P.X.W., Piroozfar, P., Wood, H., Yang, Y., Yan, L., Han, Y.u., 2019. A science
300102231301), the National Natural Science Foundation of China mapping approach based review of construction safety research. Saf. Sci. 113,
(Grant No. 72101118 and Grant No. 72071115), and the Natural Science 285–297.
Johnstone, A.H., Percival, F., 1976. Attention breaks in lectures. Educ. Chem. 13, 49–50.
Foundation of Jiangsu Province (Grant No. 164050079).
Kakhki, F.D., Freeman, S.A., Mosher, G.A., 2019. Evaluating machine learning
performance in predicting injury severity in agribusiness industries. Saf. Sci. 117,
Appendix 257–262.
Kang, K., Ryu, H., 2019. Predicting types of occupational accidents at construction sites
in Korea using random forest model. Saf. Sci. 120, 226–236.
See Table A1 Khashaba, E., El-Helaly, M., El-Gilany, A.H., Motawei, S.M., Foda, S., 2018. Risk factors
for non-fatal occupational injuries among construction workers: A case–control
References study. Toxicol. Ind. Health 34 (2), 83–90.
Kim, T., Chi, S., 2019. Accident case retrieval and analyses: Using natural language
processing in the construction industry. J. Construct. Eng. Manage. 145 (3),
Agarwal, R., Srikant, R., 1994. Fast algorithms for mining association rules. In: Proc. of 04019004. https://doi.org/10.1061/(ASCE)CO.1943-7862.0001625.
the 20th VLDB Conference, pp. 487–499. Kines, P., 2002. Construction workers’ falls through roofs: Fatal versus serious injuries.
Agrawal, R., Imieliński, T., Swami, A., 1993. Mining association rules between sets of J. Saf. Res. 33 (2), 195–208.
items in large databases, Proceedings of the 1993 ACM SIGMOD International Kirk, A., Timms, S., Rininsland, fx1., Teller, S., 2016. Data Visualization: Representing
Conference on Management of Data, pp. 207-216. Information on Modern Web. Packt Publishing Ltd.
Ajayi, S.O., Adegbenro, O.O., Alaka, H.A., Oyegoke, A.S., Manu, P.A., 2021. Addressing Lander, F., Nielsen, K.J., Lauritsen, J., 2016. Work injury trends during the last three
behavioural safety concerns on Qatari Mega projects. J. Build. Eng. 41, 102398. decades in the construction industry. Saf. Sci. 85, 60–66.
https://doi.org/10.1016/j.jobe.2021.102398. Larsson, T.J., Field, B., 2002. The distribution of occupational injury risks in the
Alsheikhly, A.S., Alsheikhly, M.S., 2018. Musculoskeletal injuries: types and Victorian construction industry. Saf. Sci. 40 (5), 439–456.
management protocols for emergency care, Essentials of Accident and Emergency Lette, A., Ambelu, A., Getahun, T., Mekonen, S., 2018. A survey of work-related injuries
Medicine. IntechOpen. among building construction workers in southwestern Ethiopia. Int. J. Ind. Ergon.
Amiri, M., Ardeshir, A., Fazel Zarandi, M.H., Soltanaghaei, E., 2016. Pattern extraction 68, 57–64.
for high-risk accidents in the construction industry: a data-mining approach. Int. J. Li, X., Yi, W., Chi, H.-L., Wang, X., Chan, A.P.C., 2018. A critical review of virtual and
Injury Control Safety Promot. 23 (3), 264–276. augmented reality (VR/AR) applications in construction safety. Autom. Constr. 86,
Amissah, J., Badu, E., Agyei-Baffour, P., Nakua, E.K., Mensah, I., 2019. Predisposing 150–162.
factors influencing occupational injury among frontline building construction Liao, C.-W., Perng, Y.-H., 2008. Data mining for occupational injuries in the Taiwan
workers in Ghana. BMC Res. Notes 12, 728. construction industry. Saf. Sci. 46 (7), 1091–1102.
Ayhan, B.U., Tokdemir, O.B., 2020. Accident Analysis for Construction Safety Using Lu, X., Davis, S., 2016. How sounds influence user safety decisions in a virtual
Latent Class Clustering and Artificial Neural Networks. J. Construct. Eng. Manage. construction simulator. Saf. Sci. 86, 184–194.
146 (3), 04019114. https://doi.org/10.1061/(ASCE)CO.1943-7862.0001762. McLeish, J., 1968. The lecture method. Cambridge Institute of Education.
Baker, H., Hallowell, M.R., Tixier, A.-P., 2020. Automatically learning construction Mistikoglu, G., Gerek, I.H., Erdis, E., Mumtaz Usmen, P.E., Cakan, H., Kazan, E.E., 2015.
injury precursors from text. Autom. Constr. 118, 103145. https://doi.org/10.1016/j. Decision tree analysis of construction fall accidents involving roofers. Expert Syst.
autcon.2020.103145. Appl. 42 (4), 2256–2263.
Berhanu, F., Gebrehiwot, M., Gizaw, Z., 2019. Workplace injury and associated factors Mučenski, V., Peško, I., Dražić, J., Ćirović, G., Trivunić, M., Bibić, D., 2015. Construction
among construction workers in Gondar town, Northwest Ethiopia. BMC workers injury risk assessment in relation to their experience and age. Proc. Eng.
Musculoskeletal Disorders 20, 523. 117, 525–533.
Bernold, L.E., Guler, N., 1993. Analysis of back injuries in construction. J. Construct. Myatt, G.J., Johnson, W.P., 2009. Making sense of data II: A practical guide to data
Eng. Manage. 119 (3), 607–621. visualization, advanced data mining methods, and applications. John Wiley & Sons.
Cheng, C.-W., Leu, S.-S., Cheng, Y.-M., Wu, T.-C., Lin, C.-C., 2012. Applying data mining Panwar, A., Jha, K.N., 2021. Integrating Quality and Safety in Construction Scheduling
techniques to explore factors contributing to occupational injuries in Taiwan’s Time-Cost Trade-Off Model. J. Construct. Eng. Manage. 147 (2), 04020160. https://
construction industry. Accid. Anal. Prev. 48, 214–222. doi.org/10.1061/(ASCE)CO.1943-7862.0001979.
Cheng, E.W.L., Li, H., Fang, D.P., Xie, F., 2004. Construction safety management: an Sánchez, A.S., Riesgo Fernández, P., Sánchez Lasheras, F., de Cos Juez, F.J., García
exploratory study from China. Construct. Innovat. 4 (4), 229–241. Nieto, P.J., 2011. Prediction of work-related accidents according to working
Chiang, Y.-H., Wong, F.-W., Liang, S., 2018. Fatal Construction Accidents in Hong Kong. conditions using support vector machines. Appl. Math. Comput. 218 (7), 3539–3552.
J. Construct. Eng. Manage. 144 (3), 04017121. https://doi.org/10.1061/(ASCE) Sarkar, S., Pramanik, A., Maiti, J., Reniers, G., 2020. Predicting and analyzing injury
CO.1943-7862.0001433. severity: A machine learning-based approach using class-imbalanced proactive and
Choi, J., Gu, B., Chin, S., Lee, J.-S., 2020. Machine learning predictive model based on reactive data. Saf. Sci. 125, 104616. https://doi.org/10.1016/j.ssci.2020.104616.
national data for fatal accidents of construction workers. Autom. Constr. 110, 1–14. Schneider, S.P., 2001. Musculoskeletal injuries in construction: a review of the literature.
Fan, C., Xiao, F.u., Yan, C., 2015. A framework for knowledge discovery in massive Appl. Occup. Environ. Hyg. 16 (11), 1056–1064.
building automation data and its application in building diagnostics. Autom. Constr. Schwatka, N.V., Butler, L.M., Rosecrance, J.R., 2012. An aging workforce and injury in
50, 81–90. the construction industry. Epidemiol. Rev. 34 (1), 156–167.
Fang, D., Jiang, Z., Zhang, M., Wang, H., 2015. An experimental method to study the Sherratt, F., Leicht, R., 2020. Unpacking ontological perspectives in CEM research:
effect of fatigue on construction workers’ safety performance. Saf. Sci. 73, 80–91. Everything is biased. J. Construct. Eng. Manage. 146 (2), 04019101. https://doi.org/
Fang, D., Zhao, C., Zhang, M., 2016. A cognitive model of construction workers’ unsafe 10.1061/(ASCE)CO.1943-7862.0001734.
behaviors. J. Construct. Eng. Manage. 142 (9), 04016039. https://doi.org/10.1061/ Singh, M., Singh, G., 2011. Development of Predictor for Sequence Derived Features from
(ASCE)CO.1943-7862.0001118. Amino Acid Sequence using Associate Rule Mining. Int. J. Comput. Sci. Security
Finneran, A., Gibb, A.G., 2013. W099: Safety and Health in construction: Research (IJCSS) 5, 14.
Roadmap report for consultation, CIB Publication 376. International Council for Siu, O.-L., Phillips, D.R., Leung, T.-W., 2003. Age differences in safety attitudes and
Research and Innovation in Building and Construction. safety performance in Hong Kong construction workers. J. Saf. Res. 34 (2), 199–205.
Garg, R., Cheung, J.P., Fung, B.K., Ip, W., 2012. Epidemiology of occupational hand Skibniewski, M., 2015. Research trends in information technology applications in
injury in Hong Kong. Hong Kong Med. J. 18, 131–136. construction safety engineering and management. Front. Eng. Manage. 1 (3), 246.
https://doi.org/10.15302/J-FEM-2014034.

15
X. Xu and P.X.W. Zou Safety Science 144 (2021) 105481

Sprinthall, R.C., Fisk, S.T., 1990. Basic statistical analysis. Prentice Hall Englewood Xiao, F.u., Fan, C., 2014. Data mining in building automation system for improving
Cliffs, NJ. building operational performance. Energy Build. 75, 109–118.
Tadesse, S., Israel, D., 2016. Occupational injuries among building construction workers Xu, S., Zou, P.X.W., Luo, H., 2018a. Impact of attitudinal ambivalence on safety
in Addis Ababa, Ethiopia. J. Occupat. Med. Toxicol. 11, 1–6. behaviour in construction. Adv. Civ. Eng. 2018, 1–12.
Tixier, A.-P., Hallowell, M.R., Rajagopalan, B., Bowman, D., 2016a. Application of Xu, X., Wang, J., Li, C.Z., Huang, W., Xia, N., 2018b. Schedule risk analysis of
machine learning to construction injury prediction. Autom. Constr. 69, 102–114. infrastructure projects: A hybrid dynamic approach. Autom. Constr. 95, 20–34.
Tixier, A.-P., Hallowell, M.R., Rajagopalan, B., Bowman, D., 2016b. Automated content Yi, W., Chan, A.P.C., 2015. Optimal work pattern for construction workers in hot
analysis for construction safety: A natural language processing system to extract weather: a case study in Hong Kong. J. Comput. Civil Eng. 29 (5), 05014009.
precursors and outcomes from unstructured injury reports. Autom. Constr. 62, https://doi.org/10.1061/(ASCE)CP.1943-5487.0000419.
45–56. Zhang, C., Zhang, S., 2003. Association rule mining: models and algorithms. Springer.
Tixier, A.-P., Hallowell, M.R., Rajagopalan, B., Bowman, D., 2017. Construction safety Zhang, F., Fleyeh, H., Wang, X., Lu, M., 2019. Construction site accident analysis using
clash detection: identifying safety incompatibilities among fundamental attributes text mining and natural language processing techniques. Autom. Constr. 99,
using data mining. Autom. Constr. 74, 39–54. 238–248.
Verma, A., Khan, S.D., Maiti, J., Krishna, O.B., 2014. Identifying patterns of safety Zhang, M., Cao, T., Zhao, X., 2017. Applying sensor-based technology to improve
related incidents in a steel plant using association rule mining of incident construction safety management. Sensors 17 (8), 1841. https://doi.org/10.3390/
investigation reports. Saf. Sci. 70, 89–98. s17081841.
Westaby, J.D., Lowe, J.K., 2005. Risk-Taking Orientation and Injury Among Youth Zhang, M., Murphy, L.A., Fang, D., Caban-Martinez, A.J., 2015. Influence of fatigue on
Workers: Examining the Social Influence of Supervisors, Coworkers, and Parents. construction workers’ physical and cognitive function. Occup. Med. 65 (3), 245–250.
J. Appl. Psychol. 90 (5), 1027–1035. Zhou, Z., Irizarry, J., Li, Q., 2013. Applying advanced technology to improve safety
Xiang, J., Bi, P., Pisaniello, D., Hansen, A., Sullivan, T., 2014. Association between high management in the construction industry: a literature review. Construct. Manage.
temperature and work-related injuries in Adelaide, South Australia, 2001–2010. Econ. 31 (6), 606–622.
Occup. Environ. Med. 71 (4), 246–252. Zou, P.X.W., Sunindijo, R.Y., 2015. Strategic safety management in construction and
engineering. Wiley Online Library.

16

You might also like