Evaluation of Machine Learning Algorithms using Feature Selection Methods for Network Intrusion Detection Systems

In this modern internet era dominated by technology and automation, network security is a concern for the whole world. Despite many security methods such as virus scanners, firewalls, encryption machines, the present day networks are vulnerable to cyber attacks and data theft. As a result, the Intrusion Detection System (IDS), with their preset blocking and filtering rules, is gaining global attention to detect and prevent the attacks in order to safeguard the network system. This paper evaluates the execution time and accuracy of different feature selection methods using different machine learning algorithms for application in IDSs. Four features selection methods (Pearson correlation, Chi-square, recursive feature elimination, tree-based: select from model) are used with three machine learning algorithms (random forest, logistic regression, k-nearest neighbor) on NSL-KDD datasets using two different machines with high and low configurations. This paper observes that, for both the machines, the best result is obtained by the random forest method in terms of execution time and accuracy.