A Hybrid Intrusion Detection System: Integrating Hybrid Feature Selection Approach with Heterogeneous Ensemble of Intelligent Classifiers

This paper proposes Hybrid Feature Selection Approach – Heterogeneous Ensemble of Intelligent Classifiers (HyFSA-HEIC) for intelligent lightweight network intrusion detection system (NIDS). The purpose is to classify for anomaly from the incoming traffic. This system hierarchically integrates HyFSA and HEIC. The HyFSA will obtain the optimal number of features and then HEIC is built using these optimal features. HyFSA helps to decrease the computation time of the system and make it lightweight to work in real time. The aim of HEIC is to obtain accurate and robust classifier and enhance overall performance of the system. The results demonstrate that proposed system outperforms other ensemble and single classifier methods used in this paper. It has true positive rate (99.9%), accuracy (99.91%), precision (99.9%), receiver operating characteristics (99.9%), low false positive rate (0.1%) and lower root mean square error rate (3.06%) with a minimum number of selected 6 features. It also reduces time to build and time to test the model by 50.79% and 55.30% respectively on reduced features set. The results evince that detection rate, accuracy and precision of the system is increased by incorporating feature selection approach with heterogeneous ensemble of intelligent classifiers and significantly reduce the computation time.