Advanced Machine Learning Techniques for Classification and Detection of Network Intrusions in Cyber-Physical Systems

Abstract

IDS keep an eye out for any indications of malicious behaviour or policy breaches in a network or system. An administrator should be notified or a central record should be kept of any intrusion activity or violation using a security information and event management system. Through the integration of sensing, computation, control, and networking, cyber physical systems connect physical infrastructure and objects to the internet and to each other. Therefore, it is necessary to have sufficient intrusion detection after safeguard the CPS from cyber-attacks, which might corrupt the equipment. CPS network intrusion categorisation and detection using sophisticated ML algorithms is the subject of this study's thorough methodology. Some of the methodical stages of the study were data collection preprocessing, and feature selection of the internet traffic that were used in the work which used NSL-KDD dataset for intrusion detection. As a result, a range of ML models is employed in classification applications that exploit their main benefits, including RF, LR, and XGBoost. This model's performance is evaluated using metrics that are based on recall, accuracy, precision, and F1 score. Confusion matrix visualisations are also included for efficient comprehension of classification results. The findings reveal that RF achieved a highest accuracy at 99.50%, followed closely by XGBoost at 99.41%, while LR recorded an accuracy of 95.39%. In addition, Kappa coefficient test, precision99.65 %, recall99.38%, and F1 score99.51 % in RF were observed. In total, the results reveal Random Forest as the superior model in comparison to employed LR and XGBoost while creating accurate IDS to protect CPS sufficiently.