Hybrid Enhanced Intrusion Detection Frameworks for Cyber-Physical Systems via Optimal Features Selection

Background/Objectives: Cyber-physical systems (CPSs) form the critical infrastructure for many nations like smart grids, home automation, smart cities, smart health care, smart automobiles, etc. These systems are susceptible to various attacks due to their wider surface area. Cyber-attacks on these systems can interrupt the critical services provided by them. Thus, intrusion detection frameworks (IDFs) are needed to identify the attacks on CPSs so that countermeasures can be taken to minimize the harm of such attacks. Limitations of existing IDFs are poor detection rate, high detection time, high false alarm rate, and large space and time complexities. The objective of this study is to design hybrid-enhanced IDFs to overcome these issues. Methods: Two enhanced IDFs are proposed in this research work. SelectKBest-MI (mutual information), framework fuses two filter-based feature selection techniques namely SelectKBest and mutual information for selecting optimal features, and the Random Forest (RF) is utilized as a classifier. The second proposed IDF is named CNN-SVM-GWO. Convolutional Neural Network (CNN) is used for extraction of attributes, Support Vector Machine (SVM) and Gray Wolf Optimizer (GWO) are used for the optimal number of feature selection, RF and Extreme Gradient Boosting (XGB) classifiers are used for intrusion detection. Two datasets have been used: CICIDS2017 and CIC-IoT-2023. Parameters considered for comparison with existing techniques are accuracy, precision, recall, F1-score, and prediction time. Findings: Implementation of SelectKBest-MI framework using the CICIDS2017 dataset, results in better accuracy of 99.99%, precision of 0.99, recall of 0.99, F1-score 0.99 for binary classification. Implementation of CNN-SVM-GWO framework using CIC-IoT-2023 dataset results in accuracy 99.60%(RF), 99.49(XGB), precision 0.99, recall 0.99, F1-score 0.99. Novelty: CNN-SVM-GWO IDF prediction time is 0.75 seconds (RF) and 0.078 seconds (XGB). The proposed model has reduced time complexity. Novel hybrid IDFs for optimal feature selection are proposed with enhanced efficiency. Keywords: Cyber-physical systems, Intrusion detection system, Optimal feature selection, Gray wolf optimizer, Convolutional neural network