Real Coded Feature Selection Integrated with Self-adaptive Differential Evolution Algorithm

Abstract

The current optimization algorithms for feature selection are mostly based on binary coded swarm intelligence algorithms. A novel real coded optimization algorithm which using the weighted distance metric is proposed in this paper, integrated with the self-adaptive differential evolution algorithm in order to self-adapting control parameter. The optimal real weight vector of all features is expected to be found to maximize the multi-class margin, and a criterion to select feature based on the optimal weight vector is given. This method is tested by classifying the breast impedance feature from UCI breast tissue dataset, and result indicates it is helpful to improve classification capability and generalized capability.