A new wrapper feature selection model for anomaly-based intrusion detection systems

摘要

Feature selection is a fundamental phase of anomaly-based intrusion detection. It is a method that selects the near-optimal subset of features to improve efficiency and reduce the number of false positives. This paper presents a new method that combines agglomerative hierarchical clustering (AHC) with a support vector machine (SVM) classifier. An intelligent process classifies features based on their variances for each attack category. Features are selected based on their variance and grouped according to their similarities. An iterative algorithm forms subsets of candidate combinations by combining the obtained attack clusters with the normal ones. The SVM classifier is applied to find the best candidate. NSL-KDD and CICIDS2017 datasets are used. The results show a significant reduction in the number of features. Moreover, it performs very well on all attacks and outperforms other existing approaches. Perfect accuracy of 100% is achieved on Heartbleed, SQL injection, and botnet attacks.