<![CDATA[Tourist arrival forecasting is a crucial aspect of planning and decision-making in the tourism sector. Accurate predictions are essential to anticipate surges or declines in visitor numbers, design effective marketing strategies, and manage resources efficiently. This study proposes a hybrid forecasting approach that integrates traditional statistical methods with machine learning algorithms to improve the accuracy of tourist arrival forecasts. Five forecasting models are implemented: ARIMA as a representative of traditional statistical models; Random Forest and Extreme Gradient Boosting (XGBoost) as machine learning models; a simple hybrid model, which combines ARIMA and XGBoost predictions through simple averaging; a weighted hybrid model, which merges the two models using performance-based weights; and a stacking hybrid model, which utilizes a meta-model to optimize prediction combinations. Given that the dataset exhibits significant pattern changes, or structural breaks, particularly during the COVID-19 pandemic, this study employs a rolling window backtesting approach for model evaluation. This method allows the models to be tested progressively across normal, crisis, and recovery periods, providing a realistic assessment of their performance under dynamic conditions. Model performance is evaluated using three key metrics: RMSE, MAE, dan MAPE. The results demonstrate that the stacking hybrid model consistently achieves the lowest RMSE across all test periods, highlighting its ability to capture complex trends and extreme fluctuations caused by COVID-19 Keywords: Rolling Window Backtesting, Weighted Hybrid, Weighted Hybrid.]]>
Copyrights © 2025
