The critical need for robust defense against cyber-attacks has driven thedevelopment of Cyber Intrusion Detection Systems (CIDSs). AlthoughCIDSs have proven good capabilities of swiftly and accurately iden-tifying malicious activities, there are still some challenges have to betackled. The exponential growth of the search space is one of thesechallenges making finding an optimal solution computationally infeasi-ble for large datasets. Furthermore, the weight space while searchingfor optimal weight is highly nonlinear. Motivated by the observedcharacteristics, complexities, and challenges in the field, this paperpresents an innovative (CIDS) named ANWO-MLP (Adaptive NonlinearWhale Optimization Multi-layer Perceptron). A novel feature selectionmethod called ANWO-FS (Adaptive Nonlinear Whale Optimization -Feature Selection) is employed in the proposed CIDS to identify themost predictive features enabling robust MLP training even in thehighly nonlinear weight spaces. The insider threat detection process is improved by investigating vital aspects of CIDS, including data pro-cessing, initiation, and output handling. We adopt ANWOA (previouslyproposed by us) to mitigate local stagnation, enable rapid conver-gence, optimize control parameters, and handle multiple objectives byinitializing the weight vector in the ANWO-MLP training with mini-mal mean square error. Experiments were conducted on various highlyimbalanced datasets to verify the robustness, stability, and efficiencyof our proposed ANWO-MLP . By employing diverse metrics, superiordetection rates are demonstrated compared to existing approaches.