An adaptive nonlinear whale optimization multi-layer perceptron cyber intrusion detection framework

The increasing prevalence of cyber threats has created a critical need for robust defense against such incidents. Many Cyber Intrusion Detection Systems (CIDSs), utilizing machine learning have been developed for this purpose. Although, these recent CIDSs have provided the capability to analyze vast amounts of data and identify malicious activities, there are still challenges to be tackled to enhance their effectiveness. The exponential growth of the search space is one of these challenges which makes finding an optimal solution computationally infeasible for large datasets. Furthermore, the weight space while searching for optimal weight is highly nonlinear. Motivated by the observed characteristics, complexities, and challenges in the field, this paper presents an innovative (CIDS) named ANWO-MLP (Adaptive Nonlinear Whale Optimization Multi-layer Perceptron). A novel feature selection method called ANWO-FS (Adaptive Nonlinear Whale Optimization-Feature Selection) is employed in the proposed CIDS to identify the most predictive features enabling robust MLP training even in the highly nonlinear weight spaces. The insider threat detection process is improved by investigating vital aspects of CIDS, including data processing, initiation, and output handling. We adopt ANWOA (previously proposed by us) to mitigate local stagnation, enable rapid convergence, optimize control parameters, and handle multiple objectives by initializing the weight vector in the ANWO-MLP training with minimal mean square error. Experiments conducted on three highly imbalanced datasets demonstrate an average efficacy rate of 98.33%. The details of the results below show the robustness, stability, and efficiency of the proposed ANWO-MLP compared to existing approaches.