A hybrid feature selection method for high-dimensional data
Nooshin Taheri, Hossein Nezamabadi–pour · 2014
Feature selection is one of the important preprocessing steps in analyzing high dimensional datasets. In this paper, first the ensemble of three different filter ranking methods including: Information Gain (IG), ReliefF and F-score are used to reduce the dimension of datasets. Afterward, reduced data are utilized as inputs of the meta-heuristic algorithm, Improved Binary Gravitational Search Algorithm (IBGSA), for selecting optimal subset of features with highest classification accuracy rate. In order to evaluate the proposed method, it is applied to several high-dimension standard datasets and the results in terms of classification accuracy and feature reduction rate are presented. The experimental results confirm the capability of the proposed algorithm.