CNN-GWO-voting & hybrid: ensemble learning inspired intrusion detection approaches for cyber-physical systems

Cyber threats are increasingly dynamic and sophisticated, often surpassing the capabilities of conventional intrusion detection systems (IDSs). Current IDSs for cyber-physical systems (CPSs) face limitations such as high computational complexity, low detection accuracy, and elevated false positive rates. To address these challenges, two novel IDS approaches inspired by ensemble learning are being proposed. The first approach, CNN-GWO-Voting, combines deep learning, evolutionary optimization, and ensemble learning to enhance effectiveness. This approach introduces a novel hybrid model design with soft voting mechanism and a novel fitness function to optimize attribute selection, distinguishing it from existing models. In this approach, a convolutional neural network (CNN) is used for feature extraction, a gray-wolf optimizer (GWO) for the selecting optimal attack relevant features, and a soft voting mechanism to blend predictions from four base classifiers: random forest, support vector machine (SVM), decision tree and XGBoost. Evaluated on the CIC-IoT-2023 dataset, this model achieves high accuracy (99.15%), precision (0.99), recall (0.99), F1-score (0.99), FPR (0.008), and FNR (0.008) while selecting only 15 of 46 features, significantly reducing computational complexity. The second approach introduces a hybrid comprehensive ensemble classifier for binary classification, utilizing logistic regression, naïve bayes, SVM, k-nearest neighbour, and multilayer perceptron as base classifiers, with ensemble technique such as voting, stacking, bagging, and boosting. On the CIC-IoT-2023 dataset, boosting yielded the best results with i.e. accuracy (98.16%), precision (0.99), recall (0.98), and F1-score (0.98). Both proposed approaches not only outperform the recent advanced techniques but also introduce significant improvements in feature extraction, optimization, and ensemble methodologies tailored specifically for the CPSs security.