Towards an intrusion detection system for detecting web attacks based on an ensemble of filter feature selection techniques

ABSTRACT The use of machine learning models in intrusion detection systems (IDSs) takes more time to build the model with many features and degrade the performance. The present paper proposes an ensemble of filter feature selection techniques (EFFST) to obtain a significant feature subset for web attack detection by selecting one-fourth split of the ranked features. The experimentation on the CICIDS 2017 dataset shows that the proposed EFFST method provides a detection rate of 99.9909%, with J48 using 24 features. The system’s performance is compared to the original features and traditional relevant feature selection methods employed in IDSs..