Reduced size multi layer perceptron neural network for human chromosome classification

In this paper we introduce a technique to reduce dimension of neural networks (NN) for classification and apply it to an improved multi layer perceptron (MLP) NN for automated classification of human chromosomes. This technique reduces number of output neurons from an order of n to log/sub 2/{n} that reduces dimension of network, number of required training data, generalization error of the network and training time significantly. All experiments of this research, including training and recall, are done using Copenhagen data set. Using 304 chromosomes for 24 classes in training mode, accuracy more than 88% is achieved in recall mode. The improved MLP training time is more than five times faster than a standard MLP. The introduced idea can be generalized to any neural network, which is used for classification.