<![CDATA[Accurate hotel tax revenue forecasting is critical for supporting proactive fiscal planning in tourism-dependent local governments . Hotel tax revenues in these regions exhibit high volatility influenced by seasonal tourism patterns, visitor preferences, economic conditions, and external shocks such as the COVID-19 pandemic . Traditional time series forecasting methods such as Autoregressive Integrated Moving Average (ARIMA) and Exponential Smoothing struggle to capture complex seasonal patterns and accommodate multiple external factors . Recent advances in time series forecasting—particularly Facebook's Prophet framework—offer automatic decomposition of trend, seasonality, and holiday effects, plus the ability to integrate external regressors . However, Prophet's performance is highly sensitive to hyperparameter configurations, and default settings often produce suboptimal results on volatile data . Bayesian Optimization has emerged as an efficient technique for hyperparameter tuning, achieving convergence with significantly fewer iterations compared to exhaustive grid search . This study develops and validates a Bayesian-Optimized Prophet Framework for forecasting monthly hotel tax revenue in Kabupaten Tana Toraja, a cultural tourism destination in Indonesia, over 60 months (January 2020–December 2024) encompassing normal conditions, pandemic disruption, and recovery phases. The optimized model achieved Mean Absolute Percentage Error (MAPE) of 9.59% compared to baseline Prophet's 33.72%—a 71.55% improvement in forecasting accuracy. Mean Absolute Error (MAE) reduced from Rp 11.76 million to Rp 3.34 million per month. Robustness testing during COVID-19 pandemic demonstrated model stability with MAPE ≤15% despite >60% revenue decline. The framework provides 24-month forecasts (2025–2026) with 95% confidence intervals and decision-support capability with lead-time advantage of 3–6 months for early revenue shortfall detection. This research contributes a reproducible, efficient methodology for hyperparameter tuning in time series forecasting within fiscal planning domain, applicable to other tourism-dependent regions and tax categories.]]>
