Feature selection of high dimensional data using Hybrid FSA-IG

Abstract

Feature selection (FS) is a process of selecting a subset of relevant features depends on the specific target variables especially when dealing with high dimensional dataset. The aim of this paper is to investigate the performance comparison of different feature selection techniques on high dimensional datasets. The techniques used are filter, wrapper and hybrid. Information gain (IG) represents the filter, Fish Swarm Algorithm (FSA) represents metaheuristics wrapper and Hybrid FSA-IG represents the hybrid technique. Five datasets with different number of features are used in these techniques. The dataset used are breast cancer, lung cancer, ovarian cancer, mixed-lineage leukaemia (MLL) and small round blue cell tumors (SRBCT). The result shown Hybrid FSA-IG managed to select least feature that represent significant feature for every dataset with improved performance of accuracy from 4.868% to 33.402% and 1.706% to 25.154% compared to IG and FSA respectively.