18th AIAI 2022, 17 - 20 June 2022, Greece

Enhanced dependency-based feature selection to improve anomaly network intrusion detection

Zakaria SAHRAOUI, Khadidja BENNACEUR, Mohamed Ahmed-Nacer


  In daily live, online computer systems are becoming more pervasive and integrated. However, the access to the Internet can produce significant issues like cyber-attacks. The network intrusion detection system (NIDS) is a promising security solution that is used to detect attacks. It recently used Deep Learning in the detection process to obtain high performance. The performance of an NIDS depends on the used training dataset and the quality of features, where irrelevant features may decrease the detection performance, oppositely to relevant ones that are able to improve it. Feature selection is a good solution to select only relevant features to participate in the detection process. Chi-square is a supervised feature selection method that select only the most dependent features of the class feature. In this work, an Enhanced Chi-square (EChi2) method is proposed to select and weight features considering its degree of relevance. Experiments results, using the well-known NSLKDD dataset, shows that the proposed method outperforms the Chi-square.  

*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.