Cement Sales Forecasting Using Backpropagation Neural Network and Recurrent Neural Network

Abstract

Backpropagation Neural Network (BPNN) is a Neural Network (NN) that moves forward and does not have a loop where the signal flow from input neurons to output neurons, while Recurrent Neural Network (RNN) is a NN model where architecture has at least one feedback loop. In this research, cement sales forecasting was carried out at PT Semen Indonesia (Persero) Tbk by using BPNN and Elman type RNN. The purpose of this research was to obtain BPNN and Elman type RNN modeling for cement sales forecasting at PT Semen Indonesia (Persero) Tbk, as well as forecasting results using the best models. The results show that the best BPNN model is the BPNN model (9-5-1) with the Levenberg-Marquardt training algorithm with Mu initialization used is 0,02 and the aktivation function used is logsig, while the best Elman type RNN model is the Elman type RNN model (9-5-1) with gradient descent with momentum and adaptive learning rate training algorithm with the momentum used is 0,2, the learning rate used is 0,2, and the activation function used is logsig. The best model for cement sales forecasting at PT Semen Indonesia (Persero) Tbk is the BPNN model (9-5-1) with forecasting result for April 2018 to December 2018.