Significance of Activation Functions in Back Propagation Neural Network:A Study on Binary Categorical Data

This work focuses on exploration of the performance measures of Back Propagation Neural Network (BPNN) learning algorithm using various activation functions and BPNN parameters viz. number of epochs, number of neurons and learning rate. The results show that the Tangent Sigmoid activation function with 20000 epochs, 15 neurons and 0.0001 learning rate has the capability to improve the accuracy of BPNN algorithm. The experiment is conducted on Diabetes dataset provided by the Department of Medicine, University of Virginia School of Medicine.