Ridgelet Probabilistic Neural Network with Genetic Algorithm selecting center vectors

To enhance the generalization performance of conventional Probabilistic Neural Network, a novel model of Probabilistic Neural Network is proposed in this paper. Ridgelet basis functions satisfying admissible condition are adopted as the activation functions in radial basis layer of the model. To reduce the size of the network and determine the parameters of ridgelet functions including the directions, locations and scales, Genetic Algorithm is utilized to select the optimal central vectors of pattern layer. Then, the normalized central vectors are taken as the direction parameters, the location parameters are all set to 1, and the scales are obtained in the training process of the network or defaulted to the same values. The classification experiments indicate that the new model has smaller size and better classification performance compared with conventional classifiers, especially in aspect of generalization performance.