A NOVEL APPROACH TO THE ACTIVATION FUNCTIONS IN DEEP NEURAL NETWORK FOR FUTURE PREDICTION
Abstract
Hidden layer plays a major role in learning process of the entire neural network model. The cautious choice of the activation function gives the best and optimal solution to the problem. In time series problem the maximum solution is gained by activation function. This paper helps to identify the best activation function for the hidden layer in order to solve the stock market forecasting. Major activation functions like linear, sigmoid, tanh and ReLU are applied into the newly developed deep learning model CDRANN. By comparing the performance result of CDRANN with the stock dataset tanh is proven to be the best activation function for the model.