The Effect of Hyper-Parameters on the Performance of Third Order Neural Network Algorithms on Medical Classification Data

Abstract

The artificial neural network (ANN) particularly back propagation (BP) algorithm has recently been applied in many areas. It is known that BP is an excellent classifier for nonlinear input and output numerical data. However, the popularity of BP comes with some drawbacks such as slow in learning and easily getting stuck in local minima. Improving training efficiency of BP algorithm is an active area of research and numerous papers have been reviewed in the literature. Furthermore, the performance of BP algorithm also highly influenced by the size of the datasets and the data preprocessing techniques that been chosen. This paper presents an improvement of BP by adjusting the two term parameters on the performance of third order neural network methods. This work also demonstrates the advantages of using preprocessing dataset in order to improve the BP convergence. The efficiency of the proposed method is verified by means of simulation on medical classification problems. The results show that the proposed implementation significantly improves the learning speed of the general back-propagation algorithm.