Variance-Index Based Feature Selection Algorithm for Network Intrusion Detection

System (NIDS). An Unsupervised Variance Indexed Feature Selection Algorithm (VIFS) is proposed and demonstrated by considering a benchmark dataset, NSL KDD. A Fast KNN classification algorithm Indexed Partial Distance Search k Nearest Neighbor(IKPDS) is applied for finding the classification accuracy. The algorithm selects a subset of features based on fitness threshold value with tolerable loss in classification accuracy but gain in computational time. In a trade of NIDS classification accuracy and computational time are two factors to decide the performance of NIDS. Feature subset length is one of the parameter to influence the computational time. So in order to evaluate the fitness value to consider the classification accuracy and feature subset length in this study. Two parameters and are related to the presence of classification accuracy quality and feature subset length. A numerical illustration presented for 0.5 9, where =[0,1] and =1-. Then identified three feature selection scenarios. Finally the merits and demerits of these three scenarios are discussed. This VIFS fulfills the gain in computational time objective with a tolerable loss in classification accuracy.