Design and Implementation of Binary Neural Network Classification Learning Algorithm
S.A. Kori · IOSR Journal of Computer Engineering · 2012
In this paper a Binary Neural Network Learning (BNN-CLA) [1] is analyzed and implemented for solving multi class problem normally the classifier are construct by combining the outputs of several binary ones.The BNNC offers high degree of parallelism in hidden layer formation for all multiple classes to reduce the time for learning.The learning method is an iterative process to optimize the classifier parameters.In this approach, overlapping problem is tackled to enhance the performance of classifier by changing hyper-sphere radius.Exhaustive testing is carried out.Accuracies and number of neuron are evaluated and compare with BNNC [4].The method have been tested on Fisher's well known Iris data data set and experimental result shown the classification ability improved by using FCLA algorithm.While comparing with BNNC[4] in most cases accuracies improved in BNNL because of elimination of samples which are lying in overlapping region of classes.Thus tacking overlapping issue improved performance of this classifier.