Enhancing Intrusion Detection Systems Through Simultaneous Feature Selection and Hyperparameter Tuning

In the realm of online security, the persistent threat of cyber-attacks, ranging from targeted incursions to widespread breaches, poses significant challenges to both the economy and society. Despite considerable efforts to counteract these threats, prevalent methods often hinge on known attack signatures and supervised learning, primarily tailored for the detection of familiar attacks. Although anomaly detection methods exist to uncover unknown attacks, recent trends have seen the emergence of hybrid approaches that seamlessly blend supervised and anomaly-based techniques. The efficacy of detecting known attacks is contingent upon finely tuned classifier settings and a robust representation of features within the data. Unfortunately, contemporary research tends to treat feature selection and hyperparameter tuning as separate processes, giving rise to inaccuracies and computational challenges. In response to this gap, our proposed approach pioneers a unified strategy for feature selection and hyperparameter tuning, leveraging optimization algorithms. Through rigorous experimentation of the NSL-KDD dataset with various supervised machine learning classifiers, we strive to identify the most effective model. We propose an approach integrating optimization algorithms, specifically the Genetic Algorithm and S-shaped Binary Whale, for simultaneous feature selection and hyperparameter tuning aiming to enhance the precision and efficiency of machine learning classifiers. The proposed approach not only promises a more holistic understanding of the interplay between feature selection and hyperparameter tuning but also strives to set a new standard in optimizing models for superior threat detection in the dynamic landscape of cybersecurity.