Modeling and Forecasting Monthly Tourist Arrivals to the United States and Indonesia Using ARIMA Hybrids of Multilayer Perceptron Models

Tourism is one of the key economic sectors contributing significantly to Gross Domestic Product (GDP) values and strengthening international relations for developed and developing countries. This study devotes more to modeling and forecasting tourist arrivals to the United States and Indonesia using ARIMA hybrids of multilayer perceptron models. The choice of using ARIMA and MLP models in forming hybrid models is to tackle linearity and nonlinearity structures respectively which exist in most of the real data. Furthermore, ARIMA and MLP models are the most powerful in reducing the RMSE values when used as an auxiliary forecasting model in modeling hybrid models even though the residuals from the main forecasting model violate distributional assumptions. Regarding individual models, the ARIMA model performed very well than the multilayer perceptron model in forecasting monthly tourist arrivals to both United States and Indonesia based on both RMSE and MAPE values. MLP-ARIMA hybrid models in which ARIMA acts as an auxiliary forecasting model are observed to better forecast monthly tourist arrivals to both United States and Indonesia than ARIMA-MLP hybrid models in which ARIMA acts as the main forecasting model, based on MAPE values. In this study, the MLP(6,1)ARIMA(0,1,1)(0,1,1)¹² hybrid model (RMSE=211,837.64 & MAPE=2.79%) and MLP(12,1)-ARIMA(0,1,1)(0,1,0)¹² hybrid model (RMSE=88,636.87 & MAPE=4.92%) are selected as the best hybrid models for forecasting monthly tourist arrivals to the United States and Indonesia, respectively.