Performance Increasing Methods for Probabilistic Neural Networks

Abstract

Through this paper, some performance increasing methods for probabilistic neural network (PNN) are presented. These methods are tested with the glass benchmark database which has an irregular class distribution. Selection of a good training dataset is one of the most important issue. Therefore, a new data selection procedure was proposed. A data replication method is applied to the rare events of the dataset. After reaching the best accuracy, a principal component analysis (PCA) is used to reduce the computational complexity of PNN. Better classification accuracy than the reference work using Bayesian EM model was achieved by PNN using these methods.