Genetic Algorithm Assisted Enhancement in Pattern Recognition Efficiency of Radial Basis Neural Network

摘要

The paper presents a feature extraction method for improving pattern classification efficiency of the radial basis function neural network. The principal component analysis in combination with preprocessing by vector autoscaling and dimensional autoscaling has been used to generate two alternate feature vector representations of the objects. A feature fusion scheme is proposed in which the two feature sets are combined by simple concatenation and then allowed to undergo genetic evolution. The fused features are obtained by applying a weighting method based on the prevalence of feature components in the terminal population. The present method of feature extraction in combination with radial basis neural network has been demonstrated to improve the classification rate for nine benchmark datasets analyzed.