Learning and validation in neural network ensembles

Abstract

Ensembles of artificial neural networks (ANN) have been used in the last years as classification/regression machines, showing improved generalization capabilities that outperform those of single networks. We propose here a simple method for learning and validation in regression/classification ensembles of ANN that leads to overtrained aggregate members with an adequate balance between accuracy and diversity. The algorithm is favorably tested against other methods recently proposed in the literature, producing an improvement in performance on the standard statistical databases used as benchmarks.