A study of artificial neural network training algorithms for classification datasets

A study of artificial neural network training algorithms for classification datasets

This paper examines the study of various feed forward back-propagation neural network training algorithms for classification problem. Performance of 11 training algorithms of neural network grouped in Gradient Descent, Resilient Backpropagation, Conjugate Gradient, Quasi-Newton, and Levenberg-Marquardt were investigated and compared using Matlab’s Neural network toolbox. Several evaluation parameters, such as number of epochs, learning rate or time, mean square of error (MSE) and accuracy (ACC), were taken into account during performance comparison of the algorithms. According to the results of this study, all training algorithms produced rather satisfactory results. Considering the results for the datasets chosen, for moderate size problems the best classification performance were obtained by the Levenberg-Marquardt, and for large-scale problems Resilient Backpropagation, Conjugate Gradient, Quasi-Newton are the best choice.