Performance Analysis of Radial Basis Function Neural Network for Pattern ClassificationDownload PDF
Feed forward neural networks with Backpropagation learning rule has been used widely for the generalized pattern classification but the ill posing and unknown local error minimum problem limits the performance of Backpropagation learning rule for the problems of large feature vectors. Another type of feed forward neural network architecture i.e. Radial Basis exhibits more efficient and general approximation with respect to Backpropagation network. The purpose of this study is to analyze the performance of Radial Basis function type feed forward neural networks for the pattern classification. Therefore to perform this analysis the task of pattern classification for hand written English vowels using radial basis function neural network is used. This Implementation has been done with the training of five different samples of hand written English vowels. Adjusting the connection strength and network parameters perform the training process in the neural network. By using a simulator program, both the algorithms i.e. BP and RBF are compared with five data sets of handwritten English language vowels. The simulated results indicate the fast & good convergence and high classification rate for the RBF network.
Keywords: Pattern Classification, Radial Basis Function Neural Network, Feed Forward neural networks, handwritten pattern Recognition