Using cooperative game theory to optimize the feature selection problem

Abstract

Feature selection is an important preprocessing step in machine learning and pattern recognition. Recent years, various information theoretic based measurements have been proposed to remove redundant and irrelevant features from high-dimensional data set as many as possible. One of the main disadvantages of existing filter feature selection methods is that they often ignore some features which have strong discriminatory power as a group but are weak as individuals. In this work, we propose a new framework for feature evaluation and weighting to optimize the performance of feature selection. The framework first introduces a cooperative game theoretic method based on Shapley value to evaluate the weight of each feature according to its influence to the intricate and intrinsic interrelation among features, and then provides the weighted features to feature selection algorithm. We also present a flexible feature selection scheme to employ any information criterion to our framework. To verify the effectiveness of our method, experimental comparisons on a set of UCI data sets are carried out using two typical classifiers. The results show that the proposed method achieves promising improvement on feature selection and classification accuracy.