Feature selection based on sensitivity using evolutionary neural network

Abstract

A novel evolutionary neural network (ENN) was presented for feature selection. In order to improve the convergent speed of the evolutionary algorithm and the accuracy of classification, a heuristic mutation operator and an adaptive mutation rate was proposed. The importance of each feature was measured with the output sensitivity, and then it was taken as the heuristic to guide the searching procedure. The evolutionary programming was employed to optimize the feature subset, and the back-propagation (BP) algorithm was used to adapt the connection weights of ENN. To evaluate the performance of the proposed approach, the experiments were conducted using three well-known classification problems. The experimental results show that the proposed method has better convergent speed and achieves fewer input features than the traditional method. Furthermore, the proposed method is superior to the traditional method in terms of training accuracy and test accuracy.