<![CDATA[Wakatobi Regency, located in Southeast Sulawesi, Indonesia, is widely recognized as one of the world’s leading marine tourism destinations due to its rich biodiversity, coral reef ecosystems, and growing ecotourism sector. Despite its international reputation, the region continues to experience substantial fluctuations in monthly tourist arrivals. These variations are influenced by seasonal tourism cycles, weather conditions, transportation accessibility, economic dynamics, and the long-term structural impacts caused by the COVID-19 pandemic. Such instability creates significant challenges for regional authorities in designing sustainable tourism policies, allocating resources, and planning infrastructure development. Therefore, accurate tourism forecasting has become increasingly important for supporting evidence-based decision making and improving destination management strategies. This study develops and evaluates a Long Short-Term Memory (LSTM)-based forecasting model to predict monthly tourist arrivals in Wakatobi using 192 monthly observations collected from January 2010 to December 2025. The proposed LSTM model is compared with three benchmark models, namely SARIMAX, XGBoost, and Transformer, to assess comparative forecasting performance. Hyperparameter optimization is conducted through a systematic grid search process, resulting in the optimal configuration consisting of a window size of 6, 128 LSTM units in the first layer, 64 units in the second layer, and a dropout rate of 0.1. Experimental results indicate that the LSTM model achieves the highest predictive accuracy with a MAPE of 8.98%, categorized as High Accuracy, alongside an MAE of 104.65 and RMSE of 143.20. These results outperform Transformer, XGBoost, and SARIMAX models. Statistical validation using the Diebold-Mariano test further confirms the superiority of LSTM forecasting performance. Additionally, the developed model is implemented in a Flask-based web application that enables regional government officials to conduct interactive forecasting with multiple prediction horizons and SHAP-based interpretability analysis.]]>
