Network Intrusion Detection Systems Design: A Machine Learning Approach
Manuel Gonçalves da Silva Neto, Danielo G. Gomes · 2019
With the increasing popularization of computer network-based technologies, security has become a daily concern, and intrusion detection systems (IDS) play an essential role in the supervision of computer networks. A current approach to detect network intrusions is the development of intrusion detection systems by employing machine learning techniques. Due to a variety of strategies used, there is a need for a systematic way that supports the decision making in a machine learning-based IDS project. In this paper, we present a systematic approach to decision-making support for algorithms selection on the IDS design. We used a very recent dataset and reduced their features from 78 to 51 through the mean decrease in impurity (MDI) feature selection technique. Afterward, we evaluated the network intrusion detection performance of eight machine learning algorithms on two dataset resampling techniques. Decision Trees, Random Forests and Multi-layer Perceptron on Stratified 10-Fold algorithms reached Precision, Recall, and F1-Scores metrics on about 98%-99% with low test times.