Forecasting domestic ship passengers in the Makassar Port using feed-forward neural network and SARIMAX

Abstract

This study aims to model and predict the number of passengers in the Makassar Port using the Feed Forward Neural Network (FFNN) and Seasonal ARIMAX (SARIMAX) models, called FFNN-SARIMAX. The SARIMAX model includes seasonal exogenous variables during Eid al-Fitr and calendar variation variables one month after Eid al-Fitr. The data is collected monthly from 2006 to 2019 and divided into 92% as in-sample and 8% as out-sample data. The ARIMAX modeling is conducted by modeling the dummy variable regression residuals using seasonal ARIMA. Meanwhile, these dummy variables and significant lags of the ACF/PACF plot from the regression model's residuals will be used as input in the FFNN model. We use 1–10 hidden neurons in the FFNN model. The accuracy of the forecast is calculated using Mean Absolute Percentage Error (MAPE). The forecast results show that the number of passengers in Makassar Port is best predicted using the ARIMAX (3,1,0)(1,0,1)12 followed by the FFNN-SARIMAX model with one hidden neuron and ARIMA (1,0,1)(1,0,0)12. The seasonal ARIMAX model can capture the big surge in ship passengers during and after the Eid Al-Fitr date.