Model Combination of Activation Functions in Neural Network Algorithms (Case: Retail State Sukuk by Group)

This study aims to maximize the activation function used in backpropogation networks in finding the best architectural model. The case study used is the sale of state retail sukuk based on professional groups. The combination of activation functions used for training and testing is tansig-tansig, tansig-purelin and tansig logsig. The architectural model used is the architectural model 6-2-1 and 6-5-1. The evaluation parameters used are epoch, MSE training, MSE testing and accuracy level of truth. Data processing is assisted by using Matlab software. The results showed that the tansig-logsig activation function had more stable results than tansig-tansig and tansig-purelin.