Pattern Classification Based on Neural Network Ensembles with Regularized Negative Correlation Learning

Abstract

In this paper, we study neural network ensembles (NNE) classifier with regularized negative correlation learning (RNCL) and its application to pattern classification. In RNCL algorithm, the regularization parameter is used to control the trade off between mean square error and regularization, and to improve the ensemble's generalization ability. We propose an automatic RNCL algorithm based on gradient descent (RNCLgd) to optimize the regularization parameter while evolving the neural network ensemble's weights. The effectiveness of the NNE classifier is demonstrated on a number of benchmark data sets. Compared with back-propagation algorithm multilayer perception (BP-MLP) classifier, it has shown that the NNE classifier with RNCLgd algorithm has better pattern classification performance.