Determination of Backpropagation Neural Network Learning Parameters and Their Effect on the Training Process

Abstract

Artificial Neural Network (ANN) is a method that has characteristics similar to human biological tissue. One of the methods used for the prediction system functions as a substitute for the human brain and nerves with the ability to learn and generalize quickly in pattern recognition. To get optimal backpropagation ANN requires appropriate input parameters in training. In this study, the training parameters for backpropagation ANN were tested by changing the value of the hidden layer and the value of the learning rate. The combination of the learning parameters in the hidden layer and the learning rate will be seen to have an effect on the time required for training, RMSE errors and the number of iterations required for each network training process. The best results from network training, it takes 12 neurons to change the hidden layer parameters. The hidden layer with a value of 12 neurons gets an RMSE error of 1.654151. The training takes 3 minutes 44 seconds. The number of iterations required is 100000 iterations. At the time of testing using the ANN architecture 2-12-1 at the number of iterations of 100000 obtained an RMSE error of 0.302868 with a training time of 18 minutes 35 seconds