CoWrap: An Approach of Feature Selection for Network Anomaly Detection

Abstract

Feature Selection (FS) is a crucial technique that picks out the most significant features from an augmented and ambiguous feature set to increase the classification accuracy of a model. As per the tremendous growth of the intrusion detection systems (IDS) research field over the past decade, now network anomaly detection can also utilize feature selection techniques to enhance performance. Several solutions for selecting the best features have been proposed, but further investigation is required to strengthen efficiency. We present a hybrid feature selection method combining the filter and the wrapper approaches for anomaly-based intrusion detection systems in this study. The proposed model offers the minimal subset of features for the highest detection accuracy using both feature to feature and feature to class correlations. Evaluated on the DDOS attack of the CICIDS2018 [1] dataset, the experiment reduced the number of features from 79 to 11, which resulted in the classification accuracy of 99.82 %.