Hybrid Cyber-Attack Detection Model on Cyber-Physical Systems Using Machine Learning Techniques

Physical objects and infrastructure can now be connected to the web and to one another using cyber-physical systems, which incorporate sensing, computation, control, and networking. Protection of these cyber-physical systems is becoming a necessity for their major stakeholders, including government agencies. Their performance affects several facets of human existence, including health, economy, and welfare. Cyber-Physical Systems (CPSs) are vulnerable to security breaches, and an attacker could potentially acquire access to the process controls for a prolonged amount of time. As a part of our research, we present a hybrid cyber-attack detection model, integrating the incremental principal component analysis (IPCA) with the two machine learning methods in parallel: k-nearest neighbor (KNN) and logistic regression (LR). Two machine learning methods with hyperparameters are combined with model adjustment to enhance their ability to predict attacks. To improve the validation accuracy, the incremental principal component analysis (IPCA) is taken into consideration. The experiments are performed using the Secure Water Treatment (SWaT) dataset, comprising details about the normal operation of the system as well as the malfunctioning of the system caused due to cyber-attacks. First, we evaluate the suggested model's efficacy in light of previous research, and it shows that out of two parallel machine learning techniques, KNN does perform exceptionally well in terms of precision attaining a value of 0.997, recall of 0.996, and F1-score of 0.996 in comparison to LR.