Ensemble Based Feature Selection Technique For Flow Based Intrusion Detection System

Abstract

The use of the internet has been drastically increasing in the last decades; hence, network security has become a big issue. To resolve these issues Intrusion detection system (IDS) plays a very important role. IDS detects malicious activity in the network by inspecting all network traffic and sending an alert message or signal to the administrator. So that the person can take the most reliable action to stop or block an attack, by using a higher number of features, we can easily develop an IDS, but it will take more computational time. Hence, it is necessary to choose a relevant set of features from the dataset, which will give us high accuracy and low computational time. In the proposed system CICIDS-2017 dataset is used, and for feature selection, the arithmetic operation is performed on Gain ratio, and Pearson's correlation. Based on the arithmetic operation result, the top 30 features are selected out of 78 features, and these reduced sets of features are fed to the machine learning classifier random forest, and results are calculated. The result shows that a reduced set of features provides better accuracy, 99.8302%, than the total 78 features.