Hybrid Metaheuristic Optimization based Feature Subset Selection with Classification Model for Intrusion Detection in Big Data Environment

Abstract

Big Data denotes an enormous set of distinct structured data attained from various heterogeneous sources stacked on the storage devices ranging from petabytes to zetabytes. Employing security in Big Data is highly a crucial process due to an exponential increase in data sizes. Therefore, intrusion detection system (IDS) is employed to detect the presence of intrusions on computers, workstations, or networks. In this view, this paper presents a Hybrid Metaheuristic Optimization based Feature Subset Selection (HMOFS) with an Optimal Wavelet Kernel Extreme Learning Machine (OWKELM) based Classification model called HMOFS-OWKELM model for IDS in big data environment. In order to handle big data, Hadoop Ecosystem is utilized. The proposed HMOFS-OWKELM model involves preprocessing to remove the unwanted noise that exists in it. In addition, the HMOFS includes the hybridization of moth flame optimization (MFO) with hill climbing (HC) based feature selection process. The HC concept is incorporated to the MFO algorithm to enhance the convergence rate. Besides, OWEKM model is applied for classification process where the optimal parameter setting in the WKELM is carried out by the rat swarm optimizer (RSO). A wide range of simulations was performed on the benchmark NSL-KDDCup dataset and the results are examined interms of different evaluation parameters. The obtained results showcased that the HMOFS-OWKELM model outperforms the other methods by offering a maximum detection accuracy of 99.67%.