<![CDATA[The tourism sector constitutes a vital component of Indonesia's economic growth, especially in Bali Province, where Ngurah Rai International Airport functions as the principal entry point for international travelers. Precise prediction of tourist arrivals is critical for strategic planning, resource distribution, and infrastructure development. Nevertheless, conventional statistical techniques often struggle to adequately capture the intricate patterns in tourism data, which exhibit both periodic regularities and non-linear characteristics shaped by external influences, including global economic fluctuations, travel regulations, and the COVID-19 pandemic. This research proposes a hybrid SARIMA-XGBoost framework that combines traditional statistical modeling with machine learning techniques to simultaneously capture linear temporal dependencies and non-linear residual patterns—an integration not previously explored for Bali's tourism forecasting. The study employs 204 monthly records of international tourist arrivals spanning January 2008 to December 2024, integrating seasonal indicators and the COVID-19 pandemic period as external covariates. The SARIMA component extracts linear temporal trends and seasonal structures, whereas XGBoost captures non-linear dynamics embedded in the residuals. The hybrid model achieves substantially higher forecasting precision with MAPE of 3.22%, MAE of 0.0492, and RMSE of 0.0597, outperforming standalone SARIMA (MAPE 25.02%, MAE 0.4305, RMSE 0.5035) and XGBoost (MAPE 7.36%, MAE 0.0736, RMSE 0.0995). These results validate that integrating statistical and machine learning methodologies significantly enhances predictive accuracy. The proposed model offers airport management, tourism boards, and policymakers a robust forecasting instrument for capacity planning and strategic decision-making, facilitating sustainable tourism development and enhancing Bali's competitiveness as an international destination.]]>
