A Comparative Analysis of Machine Learning Algorithms for Predicting Cyber Hacking Breaches in Cyber Physical Systems
V. Loganathan, N. Supriya · 2024
The cyber physical systems (CPS) are seamlessly integrated into various domains, utilising sensors to capture data, processing information, and communicating timely alerts to the relevant systems. Increasing integration also exposes these systems to potential cyber threats. Unlike faults resulting from accidents, cyber attacks are characterised by intelligent and stealthy maneuvers. Deception attacks, a subset of cyber threats, involve injecting incorrect data in different systems and compromise the cyber components to introduce incorrect information in the system. Then the system is uninformed of this attacks, may fail to detect them, and leading to performance disruptions. Therefore, adapting algorithms for the identification of such attacks is crucial. It is important to note that data generated in CPS is voluminous, diverse, and high-speed. Leveraging machine learning algorithms becomes essential for effective data analysis, pattern identification, and timely attack detection. Our proposed methodology employs the structure of machine learning algorithms for the prediction phase, promptly alerts the system to the presence of an attack during its early stages. To enhance the robustness of the system against attacks, we explore the integration of various machine learning algorithms. These algorithms are essential for enhancing the system’s detection capabilities, improving its resilience against intelligent cyber threats. Experimental analyses demonstrate that ML algorithms, including random forest, decision tree, SVM, and cat boost, outperform traditional methods in detecting attacks with higher accuracy. This multi algorithmic approach not only simplifies cyber security measures but also makes them more proactive, cost-effective, and ultimately more effective in countering cyber threats in CPS.