Explainable AI And Enhanced Krill Herd Feature Selection for Optimized Deep Learning in Intrusion Detection Systems

Optimizing ICPS attack detection deep learning models is still a challenging undertaking that calls for precise a hyperparameter adjustment. The proliferation of cyber threats necessitates robust intrusion detection technologies to fortify industrial cyber-physical systems (ICPS). The XIDS structure, developed for use in ICPS, is an improved intrusion detection system that employs Bayesian optimization methods to fine-tune the deep-learning parameters, ensemble-based filter feature selection techniques (EFFS), and explainable artificial intelligence (XAI). Combining AI that is explicable with the XAI-EKHO-FS technique, the proposed model aims to discover which attributes are critical for intrusion detection. Shapley Additive Explanations (SHAP) could potentially utilize when investigating the ways in which different variables impact the outcomes and comprehension that are anticipated. Making intrusion detection more visible, comprehensible, and flexible is just one additional way this technology improves industrial IT safety. Also, this research recommends a hybrid deep learning technique called BO-GRULSTM for precise intrusion categorization as a way to improve the performance of intrusion detection systems (IDS). To fine-tune the GRU-LSTM networks’ extremes variables, the Bayesian optimising method was used. The proposed BO-GRU-LSTM method achieves an outstanding $99.42 \%$ accuracy, surpassing GRU, LSTM, and BILSTM. Comparing this to the more antiquated GRU, LSTM, and BI-LSTM algorithms is like looking at a mountain.