Kernel selection in support vector regression for forecasting international tourist arrivals: Implications for economic resilience and national defense

Abstract

Background: International tourist arrivals are an important component of Indonesia's tourism economy and are influenced by temporal patterns and macroeconomic conditions. Accurate forecasting is therefore important to support tourism planning and economic resilience. Aims: This study aims to develop and evaluate a Support Vector Regression (SVR) model with different kernel functions for forecasting international tourist arrivals in Indonesia and to compare its performance with the ARIMAX model. Method: Monthly tourist arrival data from January 2017 to November 2024 were analyzed using SVR with linear, polynomial, radial basis function, and sigmoid kernels. Data were standardized using StandardScaler, while Grid Search was used to optimize the model parameters. Inflation and the BI Rate were incorporated as explanatory variables. The best-performing SVR model was compared with ARIMAX(1,1,1) using MSE, RMSE, MAD, and MAPE. Result: The SVR with a second-degree Polynomial Kernel achieved the best performance, with an MSE of 3,401,140,779.52, RMSE of 58,319.30, MAD of 44,565.19, and MAPE of 12.94%. The model outperformed ARIMAX(1,1,1), which obtained a MAPE of 16.23%. The results indicate that the polynomial kernel can better capture nonlinear patterns in tourist arrival dynamics. Conclusion: The SVR Polynomial model provides a useful approach for forecasting international tourist arrivals in Indonesia. Its forecasting capability can support tourism planning, resource allocation, and early identification of changes in tourism demand, contributing to more adaptive and resilient tourism management.