AI based Effective Intrusion Detection Systems

Abstract

Technologies such as Internet of Things (IoT), Cloud Computing, and Artificial intel- newlineligence (AI) have gained prominence on the mainstream internet during the past few newlineyears. This leads to an explosion in the number of connected devices and an accelerat- newlineing increase in the amount of data generated by them every day. Although technology newlineadvancements improve economic growth, but it led to a significant increase in number newlineof cyberattacks . The examination of network traffic using Intrusion Detection Systems newline(IDS) is an essential component for ensuring network security. Various techniques for newlineidentifying these types of assaults have been documented in literature. Cybercriminals newlineconstantly modify their tactics and methods to increase different traffic traces and variations in attacks. Therefore, it is challenging to develop IDS to cope with emerging vulnerabilities and zero-day attacks. newlineTo address these issues this thesis presents an empirical study to examine the effec- newlinetiveness of Machine Learning (ML) and Deep Learning (DL) algorithms in detecting newlineattacks in networks. This thesis provides various detection methods by handling the newlineissues like class imbalance, high dimensional data, reducing training time and provid- newlineing explanations. Initially a novel integrated feature extraction approach is developed to reduce feature dimensions. The objective of this research is to evaluate the influence of feature reduction on enhancement of detection accuracy, reduction in computational overhead, and improvement in the efficacy of IDS. Then, the research investigates the robustness and generalizability of ensemble-selected features by examining their per- newlineformance across various intrusion scenarios and network environments. Further, re- newlinesearch evaluates the influence of hybrid feature selection in combination with hyper newlinetuned machine learning algorithms. Then the hybrid model is analysed with imbalance newlineand balanced data. An Explainable artificial intelligence (XAI) frame work is devel- newlineoped to provide transparency in predic