Comparison of backpropagation artificial neural network and SARIMA in predicting the number of railway passengers

Abstract

Trains are one of the most popular public transportations in Indonesia. The data from the Indonesian Central Bureau of Statistics show an increasing trend in the number of train passengers in Indonesia. However, the improvement of the railway network and service is needed. This study aims to forecast the number of railway passengers in Indonesia to help the government make appropriate improvements in the railway system for the future and evaluate the potential loss due to COVID-19. We use the data from the Indonesian Central Bureau of Statistics, from January 2006 to February 2020 and assume that there is no pandemic of COVID-19. The two models we use are Backpropagation Artificial Neural Network (BPANN) and Seasonal ARIMA (SARIMA). To find the best model, we observe BPANN with various parameters and the potential SARIMA models in MATLAB and R software, respectively. Our finding is that Backpropagation Artificial Neural Network of 12-5-1 with a learning rate of 0.001 has a smaller root mean squared error (RMSE) compared to SARIMA (2,1,0)(0,1,2)12. Hence, it yields a more accurate forecast of the number of train passengers, which helps the railway company to improve and understand the loss due to COVID-19 accurately.