Towards Reliable IoT Security: A Deterministic Arithmetic Optimization Algorithm for Wrapper-Based Feature Selection in Intrusion Detection Systems

Today use computer networks all the time for everything—our phones, computers, internet of thing (IoT), and cloud services. Because of this, networks often get attacked by things like denial of service (DoS), user to remote attack (U2R) We try to stop these attacks with Intrusion Detection Systems (IDSs). However, today's IDSs struggle to find brand-new types of attacks. To make them work better, we first need to pick out only the most useful features of information before the system runs. This paper introduces a Deterministic version of the Arithmetic Optimization Algorithm (DAOA) to solve the feature selection problem in classification. The classifier employs K-Nearest Neighbors (KNN) using a wrapper-based approach. to find the optimal solutions. In contrast, all previous studies have introduced a probabilistic version of the Arithmetic Optimization Algorithm (BAOA). This study uses NF-UNSW-NB15-V2 dataset as benchmark datasets from the collection by the university of Queensland The results demonstrate that DAOA outperformed the Binary Arithmetic Optimization Algorithm(BAOA), Binary Grey Wolf Optimizer (GWO), Binary Particle Swarm Optimization (BPSO), and Binary Harmony Search optimization (HS),Binary Ant Colony Optimization for Real-valued domains (ACOR), when various performance metrics were used, including classification accuracy, selected features, The tested algorithms were ranked using the Friedman Test, and pairwise comparisons were performed using the Wilcoxon Signed-Rank Test. After running the algorithms for 30 iterations and 20 epochs, the results showed that the DAOA achieved the highest classification accuracy while selecting the smallest feature set compared to all other tested algorithms.