An optimized maximum correlation based feature reduction scheme for intrusion detection in data networks

Abstract

Wireless Networks (WNs) is a widely used technology that has found application in many fields due to their mobile and flexible nature. Many attempts have been made to secure the standard of WNs by utilizing useful security features. But due to the absence of an external robust defense mechanism such as Intrusion Detection Scheme (IDS), most of the time, the network fails to provide proper security to the application. To design an effective defense mechanism, the use of appropriate features is a must for any network. This article proposes an Optimized Maximum Correlation based Feature Reduction (OMCFR) technique for data networks. The proposed scheme utilizes maximum correlation as a major factor depending upon which individual rank is allocated to the features. The useful features are extracted using OMCFR for efficient detection. The selected features are utilized with multiclass classifier to classify the data into normal against intrusive activities. A Random Forest based multiclass classifier technique is utilized in the study. The standard dataset of Wireless Networks from the AWID family (2015), CICIDS2017 and NSL-KDD family is utilized to evaluate the proposed IDS. The results show promising performance with reduced False Positive Rate (FPR) (for NSL-KDD: 0.10, for AWID: 0.27), achieves high detection accuracy (for NSL-KDD: 99.95$\%$, for AWID: 99.2$\%$) and overall time complexity (for NSL-KDD: 182.5 s, for AWID: 812.45 s).