Machine Learning for Classification analysis of Intrusion Detection on NSL-KDD Dataset

Abstract

In the existing digital era, security concerns turn out to be a prime obstacle as it hampers the user’s privacy. Moreover, with the emergence of new technologies, enormous amount of data is present on the network which is subjected to innumerable malicious attacks and security vulnerabilities. It is therefore, essential to detect these vulnerabilities on time so that the privacy of the user’s data is not hampered and for the same intrusion detection system (IDS) is used which is deemed as the cornerstone of security. An IDS is indispensable for timely detection of cyber-attacks as it is capable of detecting intrusive activities adequately so that any potential harm to the system resources and the user base can be avoided in time. Owing to their understanding of IDS for reducing security threats, in the current work, machine learning classifiers (MLC) were used for classifying the data. The system performance was evaluated using four diverse attribute subsets obtained from NSL-KDD dataset. For optimization, prior to the training/testing phase, the dataset was pre-processed so that irrelevant features could be removed as their contribution is inconsequential in detecting attack classes. Finally, the overall model accuracy for different attack classes namely DoS, Probe, U2L, and R2L was compared to detect the most suitable algorithm for a particular attack class.