A Feature Selection Based DNN for Intrusion Detection System

The goal of networking has the idea of “resource sharing” and “communication” in a convenient way. However, more convenience services are provided, more problems of security and privacy issues may occur. In order to prevent these problems, an IDS (Intrusion Detection System) is designed to enhance the network security and to observe abnormal behavior. Model accuracy and the training time required to build the model are affected greatly if we use the unselected features and irrelevant data. This is the reason why the selection of features is a significant process in building an Intrusion Detection System (IDS). This paper aims to boost the Deep Neural Network (DNN) capabilities by selecting the feasible features before processing networking data. This research employed the KDD Cup 99 dataset which is considered as one of the representative datasets for intrusion detection. Based on our experimental results, it is concluded that the selection of the proper features has effects on the improvement of IDS compared to the method without feature selection. This research has proved that the improvement of DNN for IDS can reach up to 99.4% for accuracy, 99.7% for precision, 97.9% for recall, and 98.8 for F1 score.