Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Threat Detection in Cyber Security by Machine

Learning
Anil Kumar Sharma Abhishek Suman Abhijeet Shukla Prof. Rajeshwari k
Dept. of Information Science Dept. of Information Science Dept. of Information Science Dept. of Information Science
BMS College of Engineering BMS College of Engineering BMS College of Engineering BMS College of Engineering
Bengaluru, India Bengaluru, India Bengaluru, India Bengaluru, India
anilsharma.is19@bmsce.ac.in abhisheksuman.is19@bmsce.ac.in abhijeet.is19@bmsce.ac.in rajeshwarik.ise@bmsce.ac.in

Abstract — In the rapidly evolving field of cybersecurity, the


students, and to conduct research and analysis. They can
also be used by traders and individual investors as a way to
ability to detect and mitigate potential threats is paramount.
test their investment strategies, without the risk of losing
With the increasing amount of data being generated by real money.
networked devices, traditional security methods are no longer
sufficient to keep networks secure. Machine learning models These platforms have a broad range of uses, including
have the potential to provide a solution by enabling effective those for education and entertainment. Virtual stock
threat detection in real-time. This project proposes the use of a trading platforms provide a risk-free opportunity for
people who are new to stock trading to become familiar
machine learning model for cyber threat detection. The model
with the stock market, grasp how trading works, and create
uses a combination of feature engineering architecture to analyze
a trading strategy. Virtual stock trading platforms give
network traffic data and detect suspicious activity. seasoned traders a risk-free way to test out new trading
concepts and tactics. These platforms furthermore give
financial firms and educators a platform to instruct clients
INTRODUCTION
and pupils about stock trading. In general, virtual stock
A cybersecurity threat is the threat of a malicious attack by an trading platforms give people a flexible and accessible
individual or organization attempting to gain access to a instrument to develop their stock trading expertise.
network, corrupt data, or steal confidential information. No
organization is immune from cyberattacks and data breaches. Virtual stock trading platforms expand their potential as a
Some cyberattacks can even destroy computer systems. useful resource for people wanting to advance their stock
trading knowledge and abilities by offering instructional
Use AI and ML to detect threats. Machine learning works. It resources. Users can learn about different facets of stock
uses algorithms to process large amounts of ever-changing trading, like market analysis, risk management, and trading
data. In cybersecurity, this means we have increasingly methods, through these tutorials and resources. Articles,
sophisticated tools to detect patterns, predict threats, and videos, webinars, and even online courses are examples of
leverage up-to-the-second information. In the rapidly these educational resources. These learning resources are
evolving field of cybersecurity, the ability to detect and designed to arm users with the information and expertise
mitigate potential threats is paramount. With the increasing they need to make wise judgements when trading stocks in
amount of data being generated by networked devices, the real world.
traditional security methods are no longer sufficient to keep
networks secure. Machine learning models have the potential BACKGROUND
to provide a solution by enabling effective threat detection in
real-time. This project proposes the use of a machine learning Threat Detection in Cybersecurity using Machine Learning
model for cyber threat detection. The model uses a With the rapid growth of digital systems and networks, the
combination of feature engineering architecture to analyze threat landscape in cyberspace has become increasingly
network traffic data and detect suspicious activity. complex and challenging to mitigate. Traditional security
approaches that rely on rule-based systems and signature-based
The objective of a cyber security detection platform is to
detection methods are struggling to keep pace with the
provide users with a simulated environment in which they can
ever-evolving nature of cyber threats. As a result, there is a
buy and sell stocks, learn about the stock market, and practice
growing need for more advanced and adaptive techniques to
investment strategies. This is achieved by offering a variety of
detect and mitigate these threats effectively.
features, such as real-time stock quotes, charts, market news,
educational resources, and contests or competitions. The
Machine learning (ML) has emerged as a powerful tool in the
ultimate goal is to help users develop their knowledge and
field of cybersecurity due to its ability to analyze vast amounts
skills in investing, and to provide them with a hands-on
of data and identify patterns that indicate malicious activities.
experience that can prepare them for real-world investing.
By leveraging ML algorithms, security systems can learn from
In addition, virtual stock trading platforms can also serve as a historical data, adapt to new threats, and continuously improve
tool for financial institutions to train employees and their detection capabilities.
technique for insider threat detection based on classifying user
Threat Detection in Cybersecurity using Machine Learning
activities in time series. Initially, a set of one-day features is
With the rapid growth of digital systems and networks, the computed from the user activity logs. These features capture
threat landscape in cyberspace has become increasingly relevant information about the activities. Subsequently, a time
complex and challenging to mitigate. Traditional security series feature vector is generated by analyzing the statistics of
approaches that rely on rule-based systems and signature-based each one-day feature over a specific period of time. The labels
detection methods are struggling to keep pace with the for each time series feature vector, indicating whether they are
ever-evolving nature of cyber threats. As a result, there is a malicious or non-malicious, are extracted from the ground truth
growing need for more advanced and adaptive techniques to data.
detect and mitigate these threats effectively. To effectively classify the unbalanced insider threat data, which
Machine learning (ML) has emerged as a powerful tool in the typically contains only a small number of malicious instances,
field of cybersecurity due to its ability to analyze vast amounts a cost-sensitive data adaptation technique is employed. This
of data and identify patterns that indicate malicious activities. technique involves randomly sub-selecting instances from the
By leveraging ML algorithms, security systems can learn from non-malicious class. As the classifier, a two-layer deep neural
historical data, adapt to new threats, and continuously improve autoencoder network is utilized and compared against other
their detection capabilities. commonly used classifiers such as random forest and
Cybersecurity is a very important part of all businesses. After multilayer perceptron.
all, if a hacker manages to penetrate their systems, they are The proposed approach is evaluated using the CMU Insider
toast! The most difficult component of cybersecurity is Threat Data, the sole publicly available insider threat dataset
figuring out if the connection requests to the system are comprising web browsing logs, login records, device
legitimate and if suspicious looking activities like receiving connections, file transfers, and email logs, totaling
and sending large amounts of data are coming from approximately 14 GB of data. The evaluation demonstrates that
professionals within the company or from cyber threats. This is both the deep autoencoder and random forest classifiers
very difficult for cybersecurity professionals to detect, achieve high precision, recall, and f-score when classifying the
especially in large enterprises where requests are constantly in time series feature set. Although the multilayer perceptron
the thousands and humans are not always accurate. This is exhibits high recall, its precision and f-score are comparatively
where machine learning can be of great help to professionals. lower than the other two classifiers.
A cyber threat detection system based on AI and ML can be
used to monitor all outbound and inbound calls, as well as all
requests to the system, to monitor for suspicious activity.
Versive, for example, is an artificial intelligence provider that
offers cybersecurity software combined with AI.
v
I. LITERATURE SURVEY
With the rapid increase in Internet traffic, the importance of
cybersecurity is also significantly growing. Areas such as IP
traffic classification, intrusion detection, spam, and malware
require attention. Traditional techniques alone are insufficient
in addressing these cybersecurity challenges, and it is
necessary to adapt to ongoing changes to maintain up-to-date
security measures.
Machine learning has emerged as a valuable tool in the field of
cybersecurity, augmenting traditional rule-based algorithms
and enhancing their efficiency through artificial intelligence
integration. While achieving full automation of analysis and
detection remains an appealing objective, significant
improvements can be made in essential aspects of Fig 1. Architecture of virtual stock trading system.
cybersecurity. This paper aims to explore the utilization of
diverse machine learning algorithms to tackle prevalent B.Using Convolutional Neural Networks to
cybersecurity issues. Network Intrusion Detection for Cyber
A. Use of Machine Learning algorithms for Threats
designing efficient cyber security solutions Defenders require an efficient network detection
The scenario-based insider threat detection focuses on approach that can quickly learn and detect new
identifying and mitigating the risks associated with malicious network behavioral characteristics to identify
activities originating from within an organization. These threats network intrusions. Deep learning techniques,
can arise from intentional or unintentional misuse of the known for their ability to outperform traditional
organization's systems, networks, data, and resources. approaches in various applications, have shown
Detecting insider threats poses a challenge as it involves promise in this regard.
individuals who have authorized access and are trusted partners This study focuses on network intrusion detection
within the organization. using Convolutional Neural Networks (CNNs) based
Current research in insider threat detection primarily revolves on the LeNet-5 architecture to classify network
around developing unsupervised behavioral anomaly detection threats. The experimental results demonstrate a high
techniques. The goal is to detect anomalies or abnormal prediction accuracy of up to 99.65% for intrusion
changes in user behavior over time. However, it is important to detection, based on a dataset of over 10,000 samples,
note that anomalous activity does not always indicate malicious with an overall accuracy of 97.53%.
intent and can potentially lead to an insider threat scenario. Many existing approaches to cyber attack detection
In order to improve existing approaches, this paper proposes a involve cyber threat analysis, aiming to identify
Malicious attackers often attempt to bypass firewalls A framework to evaluate the use of Machine Learning and
and virus detection engines by utilizing legitimate Natural Language Processing technologies for disease
network protocols such as HTTP, ICMP, and SSL. They prediction systems is proposed. A dataset of
may download malicious applications from users to disease-symptom information from a reputable UK National
remote hosts controlled by command-and-control Health Service website was collected and data cleaning,
servers. Log analysis alone proves challenging in pre-processing, and NLP techniques to extract symptom
distinguishing between legitimate and malicious descriptions were performed. The Python 'nltk' library was
connections. Additionally, qualitative approaches, such used for pre-processing methods such as capitalization, stop
as information-gathering, help detect traffic anomalies word removal, tokenization, and stemming. The data using
and differentiate benign from malicious connections in word clouds and clustering algorithms such as k-mean and
the presence of unknown threats. However, they Latent Dirichlet Allocation was also evaluated. The goal of
struggle to accurately classify threats in response to this framework is to demonstrate the potential for using
cyberattacks, leading to the problem of false positive NLP in disease prediction systems.
rates when detecting malicious connections using data The study describes the use of a Multinomial Naive Bayes
flow analysis. (MNB) method for probability classification in a prediction
To enhance the classification accuracy in threat component. The method involves two steps: first,
detection and reduce the false positive rate for Network calculating the prior probability of each class in the training
Intrusion Detection Systems (NIDS), this study dataset, and second, determining the likelihood of a
introduces an improved behavior-based classification symptom occurring in a particular class. These calculations
learning model using CNNs. The CNN is utilized to allow the classification of new unseen data.
extract enhanced behavioral features and employ The study highlights several challenges with current
softmax activation to identify the class of threats, using technologies related to prediction. One issue is the problem
collected statistical data. The study utilizes network of undetermined prediction, where multiple suggestions are
intrusion information flows from network nodes in the provided with the same probability scores when there are
National Centre for High Performance Computing only a few symptoms provided. Another issue is the lack of
(NCHC) and employs clustering analysis to categorize transparency in the black box
the behavioral characteristics of different IP addresses
for comparison. Feature vectors are then transformed C.Cyber Threat Detection based on
into feature matrices, which are treated as input images ArtificialNeural Networks using Event Profiles
for the CNN. This approach accurately categorizes
cyber threats based on the collected behavioral Detecting cyber threats in an automated and effective
characteristics derived from network traffic packet manner is a major challenge in the field of cybersecurity. In
analysis, enabling the defense system to respond this paper, we propose an artificial intelligence (AI)
promptly to high-risk threats. technique for cyber threat detection based on artificial
The rest of this article is organized as follows: Section II neural networks (ANNs). Our approach involves converting
reviews previous studies on CNNs, Section III presents a large volume of collected security events into individual
the proposed approach for a network intrusion detection event profiles and utilizing deep learning methods for
model with a CNN architecture in an online information enhanced cyber threat detection. To accomplish this, we
security management system. The experimental results developed an AI-based Security Information and Event
are discussed in Section VI, and finally, Section V Management (AI-SIEM) system that leverages event
concludes the paper. profiles for data preprocessing and employs various types of
artificial neural networks, including fully connected neural
networks (FCNNs), convolutional neural networks (CNNs),
and long short-term memory networks (LSTMs).
components of the system by exchanging Intrusion
D.Studying machine learning techniques for Detection Message Exchange Format (IDMEF) messages.
intrusion detection systems These messages provide information about the incident
source, time of occurrence, and a classification of the
This survey paper focuses on providing a comprehensive
alarm.
review of data mining and machine learning methods
applied in the context of cyber security intrusion By utilizing the OCSVM-based intrusion detection
detection. The authors conduct a literature review, module, the paper aims to enhance the security of
identifying relevant papers based on their citation count SCADA systems by detecting and responding to
and relevance to emerging methods. For each method, malicious network traffic promptly. The IDS architecture
they provide brief instructional descriptions to give developed as part of the CockpitCI project enables
readers an understanding of the approach. Since data efficient communication and collaboration between
plays a crucial role in ML and DM approaches, the paper various components involved in the detection and
also describes well-known cyber data sets commonly response process.
used in ML and DM research for cybersecurity. The
complexity of ML and DM algorithms is discussed, The increasing number of cyberattacks has highlighted
highlighting the challenges associated with their the need for improved Intrusion Detection Systems (IDS).
application in the cybersecurity domain. The paper also Machine learning techniques (ML) have emerged as
offers recommendations on when to utilize specific essential tools for early attack classification within IDS.
methods based on the characteristics of the problem at However, with numerous available algorithms, selecting
hand. By providing a survey of data mining and machine the most suitable method can be challenging. To address
learning methods for cyber security intrusion detection, this issue, this paper analyzes state-of-the-art intrusion
this paper serves as a valuable resource for researchers detection methods, discussing their advantages and
and practitioners in the field. It offers insights into the disadvantages. It also provides an overview of different
different approaches and data sets available, helping ML methods, identifying four approaches as the most
readers navigate the complexities and challenges suitable for attack classification.
associated with applying ML and DM techniques to The study evaluates the performance of IDS using several
enhance cybersecurity. selected algorithms. The IDS categorizes binary and
multiclass attacks by determining whether the traffic is
E.Intrusion Detection in SCADA systems using benign or malicious. Experimental results demonstrate
Machine Learning Techniques that binary classification consistently achieves high
This paper introduces an intrusion detection module accuracy, ranging from 0.9938 to 0.9977, while multiclass
specifically designed for detecting malicious network classification ranges from 0.9294 to 0.9983. Among the
traffic in Supervisory Control and Data Acquisition algorithms, multiclass classification with the k-nearest
(SCADA) systems. Malicious data can disrupt the normal neighbor algorithm achieves the highest accuracy of
functioning of SCADA systems and impact their 0.9983, while Random Forest achieves the highest score
operations. To address this issue, the paper proposes the of 0.9977 in binary classification. These results indicate
use of an intrusion detection mechanism called One-Class that multiclass classification performs better in intrusion
Support Vector Machine (OCSVM). Unlike other detection by accurately distinguishing between different
techniques, OCSVM does not require labeled training types of attacks and enabling a more targeted response.
data or prior knowledge of the expected anomalies,
making it suitable for processing SCADA environmental
data and automating SCADA performance monitoring.
The OCSVM module is trained using network traces and
is capable of detecting anomalies in real time within the
SCADA system. It is implemented as part of an Intrusion
Detection System (IDS) developed within the CockpitCI
project. The module communicates with other
II. PROPOSED SYSTEM parameters to optimize their performance. Employ
techniques like cross-validation and hyperparameter
Proposing a system for threat detection in cybersecurity tuning to ensure robustness and generalization of the
using machine learning involves several components and models.
considerations. Here's an outline of a potential system
architecture: Model Evaluation: Evaluate the trained models using the
validation set to assess their performance metrics, such as
Data Collection: Gather a comprehensive dataset accuracy, precision, recall, and F1-score. Adjust the
consisting of various cybersecurity-related data sources, models or experiment with different algorithms as
such as network logs, system logs, firewall logs, needed to improve their performance.
intrusion detection system (IDS) alerts, and threat
intelligence feeds. This dataset will serve as the Real-Time Monitoring: Deploy the trained model in a
foundation for training and testing the machine learning real-time monitoring system that continuously processes
models. incoming data streams and flags potential threats or
anomalies. This system should be capable of handling
Data Preprocessing: Cleanse and preprocess the collected high-volume, high-velocity data to detect and respond to
data to ensure its quality and compatibility with the threats promptly.
machine learning algorithms. This step may involve data
normalization, handling missing values, removing noise, Alert Generation and Response: When a potential threat
and feature engineering to extract relevant information is detected, generate alerts or notifications to relevant
from the raw data. personnel or security systems. These alerts should
include detailed information about the detected threat and
Feature Selection/Extraction: Identify the most relevant recommended actions to mitigate or investigate the
features from the preprocessed data that are likely to incident further.
have discriminatory power in distinguishing between
normal and malicious activities. This step helps reduce Continuous Improvement: Continuously update and
the dimensionality of the dataset and improve the retrain the machine learning models to adapt to evolving
efficiency and effectiveness of the machine learning threats and changing attack patterns. Regularly
algorithms. incorporate new data and threat intelligence feeds to
enhance the system's accuracy and effectiveness over
Model Selection: Choose appropriate machine learning time.
algorithms for threat detection based on the
characteristics of the problem and the available data. It's important to note that building an effective threat
Commonly used algorithms for cybersecurity threat detection system requires a multidimensional approach
detection include anomaly detection techniques (e.g., that combines machine learning with other cybersecurity
Isolation Forest, One-Class SVM) and supervised techniques, such as signature-based detection, behavioral
learning algorithms (e.g., Random Forest, Support Vector analysis, and human expertise. Additionally, regular
Machines). monitoring, maintenance, and updating of the system are
necessary to ensure its ongoing effectiveness in detecting
Model Training: Split the preprocessed dataset into emerging threats.
training and validation sets. Use the training set to train
the selected machine learning models, fine-tuning their
educated decisions, and obtain a deeper grasp of the V. CONCLUSION
complexity of the stock market. In conclusion, the application of machine learning algorithms
to cybersecurity has brought significant advances in threat
III. METHODOLOGY detection. The ability to analyze vast amounts of data and
The approach used in this research report includes the identify patterns that could indicate a security breach has
development of a threats detection platform using multiple made machine learning a powerful tool for organizations.
technologies and frameworks. The necessary libraries, such as However, developing these models is not without its
Pandas, Seaborn, and Matplotlib, are imported.The dataset is challenges. One of the biggest challenges is the need for
loaded from a CSV file using Pandas' read_csv function.Data high-quality, relevant and diverse training datasets. Inaccurate
cleaning and preprocessing are performed:The "Time" or biased datasets can lead to false predictions, resulting in
column is split into "Start time" and "Last time" false alarms or missed threats. Another challenge is the need
columns.Unnecessary columns are dropped from the dataset. for machine learning models to be transparent and
Missing values in the "Attack subcategory" column are filled
explainable. This is important to build trust among
with "Not Registered". Duplicated rows are removed from the
stakeholders and to understand the rationale behind the
dataset.Invalid values for source and destination ports are
filtered out.The values in the "Protocol" and "Attack model's predictions. Despite these challenges, the potential
category" columns are standardized. Exploratory Data benefits of machine learning for threat detection are too great
Analysis (EDA) is conducted:Missing values are visualized to ignore. Machine learning can significantly reduce the time
using the missing no library.The number of unique protocols and resources required to detect and respond to threats,
and attack categories is calculated.Bar charts are created to improve the accuracy of threat detection, and help
show the count and distribution of attack categories. organizations stay ahead of evolving cyber threats. To fully
realize the potential of machine learning for cybersecurity,
The code provided generates visualizations such as bar plots,
scatter plots, heatmaps, and violin plots to analyze various organizations must invest in developing robust and reliable
aspects of the dataset. It also calculates statistical measures machine learning models and continuously monitor and
like correlation coefficients and performs hypothesis testing improve these models over time. In conclusion, machine
using the t-test.The results and visualizations obtained from learning is a promising technology for cybersecurity threat
this code can be used to gain insights into the cybersecurity detection that has already proven its effectiveness in many
attacks dataset and understand patterns, trends, and real-world applications. The ongoing development and
relationships between different variables. refinement of machine learning algorithms will likely lead to
even more advanced and powerful threat detection systems in
the future, which will play a critical role in protecting
IV. RESULTS organizations systems and data from cyber threats.

The results of threat detection in cybersecurity using


machine learning can include improved accuracy in VI. ACKNOWLEDGEMENTS
distinguishing between normal and malicious activities,
early detection of threats, scalability for processing large We extend our heartfelt gratitude to Dr. B. S. Ragini Narayan,
volumes of data, adaptability to new threats, reduced Dr. P. Dayananda Pai, and Dr. S. Muralidhara for their
response time, enhanced incident investigation, and unwavering support and the provision of necessary
optimized resource allocation. While no system is infrastructure, which played a crucial role in the successful
foolproof, machine learning can significantly enhance completion of this Capstone Project Phase-1. Their support
threat detection capabilities, leading to proactive and resources were instrumental in our project's progress and
mitigation measures and faster response to security ultimate success.
incidents. Continuous monitoring and improvement are
necessary to adapt to evolving threats and maintain Our deepest appreciation goes to Dr. Jayarekha P, the Head of
effectiveness over time. the Department of Information Science and Engineering,
along with Dr. Nalini M K and Prof. Harini S, the Project
Coordinators, for their constant guidance and support
throughout the project. Their expertise, valuable insights, and VII. REFERENCES
unwavering dedication have significantly contributed to
shaping our research and ensuring its quality. [1] Soni, S. and Bhushan, B., 2019, July. Use of Machine Learning
algorithms for designing efficient cyber security solutions. In 2019
Special thanks are extended to Prof. Rajeshwari k, our 2nd International Conference on Intelligent Computing,
dedicated guide and Assistant Professor in the Department of Instrumentation and Control Technologies (ICICICT) (Vol. 1, pp.
Information Science and Engineering. We are sincerely 1496-1501). IEEE.
grateful for her continuous support, timely guidance, and [2] Chen, D., Wawrzynski, P. and Lv, Z., 2021. Cyber security in smart
cities: a review of deep learning-based applications and case studies.
mentorship. Her expertise and encouragement have been Sustainable Cities and Society, 66, p.102655.
invaluable in navigating the challenges of the project and [3] Chattopadhyay, P., Wang, L. and Tan, Y.P., 2018. Scenario-based
achieving its objectives. insider threat detection from cyber activities. IEEE Transactions on
Computational Social Systems, 5(3), pp.660-675.
We would also like to express our gratitude to all the [4] Lin, W.H., Lin, H.C., Wang, P., Wu, B.H. and Tsai, J.Y., 2018, April.
Using convolutional neural networks to network intrusion detection
faculty members of the Department of Information Science for cyber threats. In 2018 IEEE International Conference on Applied
and Engineering for their contributions, including their System Invention (ICASI) (pp.1107-1110). IEEE.
insightful views and encouraging ideas. Their support and [5] Lee, J., Kim, J., Kim, I. and Han, K., 2019. Cyber threat detection
collaboration have greatly enriched our project and based on artificial neural networks using event profiles. Ieee Access,
fostered a conducive learning environment. 7, pp.165607-165626.
[6] Ganne, A., IoT Threats & Implementation of AI/ML to Address
Emerging Cyber Security Issues in IoT with Cloud Computing.
[7] Dang, Q.V., 2019. Studying machine learning techniques for
intrusion detection systems. In Future Data and Security
Engineering: 6th International Conference, FDSE 2019, Nha Trang
City, Vietnam, November 27–29, 2019, Proceedings 6 (pp.
411-426). Springer International Publishing.
[8] Maglaras, L.A. and Jiang, J., 2014, August. Intrusion detection in
SCADA systems using machine learning techniques. In 2014
Science and Information Conference (pp. 626-631). IEEE.
[9] Bindra, N. and Sood, M., 2019. Detecting DDoS attacks using
machine learning techniques and contemporary intrusion detection
dataset. Automatic Control and Computer Sciences, 53, pp.419-428.
[10] Allard, F., Dubois, R., Gompel, P. and Morel, M., 2011. Tunneling
activities detection using machine learning techniques. Journal of
[11] Ullah, F., Naeem, H., Jabbar, S., Khalid, S., Latif, M.A.,
Al-Turjman, F. and Mostarda, L., 2019. Cyber security threats
detection in the internet of things using a deep learning approach.
IEEE access, 7, pp.124379-124389.
[12] Dalal, K.R. and Rele, M., 2018, October. Cyber Security: Threat
Detection Model based on Machine learning Algorithm. In 2018 3rd
International Conference on Communication and Electronics
Systems (ICCES) (pp. 239-243). IEEE.
[13] Joloudari, J.H., Haderbadi, M., Mashmool, A., GhasemiGol, M.,
Band, S.S. and Mosavi, A., 2020. Early detection of the advanced
persistent threat attack using performance analysis of deep learning.
IEEE Access, 8, pp.186125-186137.
[14] Buczak, A.L. and Guven, E., 2015. A survey of data mining and
machine learning methods for cyber security intrusion detection.
IEEE Communications surveys & tutorials, 18(2), pp.1153-1176.
[15] Liu, Q., Li, P., Zhao, W., Cai, W., Yu, S. and Leung, V.C., 2018. A
survey on security threats and defensive techniques of machine
learning: A data driven view. IEEE access, 6, pp.12103-12117.

You might also like