Improvisation of Training Algorithm through Hybridization for Rule Extraction

Absrtact The Artificial Neural Network is widely used for classification. In classification, training and learning of the network in which weights and biases of the network neuron are computed to give expected output, is a complex task. In this paper, we propose hybridization of back propagation and LevenbergMarquardt training algorithms. The gradient derivative with respect to the weight of the network of the gradient descent algorithm is used in augmenting Hessian matrix of Levenberg Marquardt training algorithm to update the weight and bias of the network to converge to output. The hybrid algorithm is experimented on two data sets. Experimental results show that it helps to achieve better network performance and extracts fewer rules.