Empirical Analysis of Machine Learning Models towards Adaptive Network Intrusion Detection Systems

The exponential growth of Internet access has resulted in a huge influx of network attacks, some of which are lethal or have devastating results. As a consequence, the plethora of new threats revealed makes it harder to provide network security in the direction of identifying breaches. Furthermore, the intruders with intent launching various physical attacks within the network also cannot be overlooked. An Intrusion Detection System (IDS) is a tool which inspects internet traffic for the purposes of ensuring its confidentiality, integrity, and availability as well as defending the network from potential intrusions. Despite the researchers' best efforts, IDS still poses potential challenges and it has improved accuracy and reducing false alarm rates, and recognized new intrusions. Machine learning (ML) and deep learning based IDSs have recently been adopted as viable solutions for swiftly identifying suspicious breaches. This review paper tries to present a deep brief on various IDS concepts, evaluation metrics, and dataset selection measures. This paper is carried out by carefully considering the various ML algorithms applied in the IDS domain and also tries to figure out various results obtained by using various ML algorithms towards in field of NIDS. This paper addresses the obstacles and future potential in the field of IDSs by assessing existing illustrative research. The outcomes of several studies were examined and contrasted, providing a clear path and roadmap for future research. The extensive review of recently developed IDS models and a detailed comparative analysis show the novelty of the work. The proposed paper gives an outline of different endeavors to foster an effective IDS using single, hybrid, and ensemble ML classifiers, as well as the results of these efforts using various datasets.