Developing Cybersecurity Systems Based on Machine Learning and Deep Learning Algorithms for Protecting Food Security Systems: Industrial Control Systems

Abstract

Industrial control systems (ICSs) for critical infrastructure are extensively utilized to provide the fundamental functions of society and are frequently employed in critical infrastructure. Therefore, security of these systems from cyberattacks is essential. Over the years, several proposals have been made for various types of cyberattack detection systems, with each concept using a distinct set of processes and methodologies. However, there is a substantial void in the literature regarding approaches for detecting cyberattacks in ICSs. Identifying cyberattacks in ICSs is the primary aim of this proposed research. Anomaly detection in ICSs based on an artificial intelligence algorithm is presented. The methodology is intended to serve as a guideline for future research in this area. On the one hand, machine learning includes logistic regression, k-nearest neighbors (KNN), linear discriminant analysis (LDA), and decision tree (DT) algorithms, deep learning long short-term memory (LSTM), and the convolution neural network and long short-term memory (CNN-LSTM) network to detect ICS malicious attacks. The proposed algorithms were examined using real ICS datasets from the industrial partners Necon Automation and International Islamic University Malaysia (IIUM). There were three types of attacks: man-in-the-middle (mitm) attack, web-server access attack, and telnet attack, as well as normal. The proposed system was developed in two stages: binary classification and multiclass classification. The binary classification detected the malware as normal or attacks and the multiclass classification was used for detecting all individual attacks. The KNN and DT algorithms achieved superior accuracy (100%) in binary classification and multiclass classification. Moreover, a sensitivity analysis method was presented to predict the error between the target and prediction values. The sensitivity analysis results showed that the KNN and DT algorithms achieved R2 = 100% in both stages. The obtained results were compared with existing systems; the proposed algorithms outperformed existing systems.