Interpretable AI for Hotel Occupancy Forecasting in Taiwan’s Tourist Hotels: Lessons Learned from Multiple Periods of Crisis and Market Transition

The economic downturn (e.g., the COVID-19 pandemic) caused unprecedented disruption to the global hospitality industry, fundamentally reshaping hotel occupancy patterns and demand structures. This study examines how hotel occupancy dynamics in Taiwan evolved across the pre-COVID (2017–2019), COVID (2020–2022), and post-COVID recovery (2023–2024) periods using machine learning and explainable artificial intelligence (AI) techniques. We analyze shifts in the distribution of occupancy categories, changes in the relative importance of demand-side predictors—particularly domestic and international visitors—and the role of operational and employment-related features in supporting occupancy resilience. Using proprietary Taiwanese hotel data, this study classifies occupancy levels into three classes: low (<50%), medium (50–75%), and high (>75%). Based on these classifications, separate Gradient Boosting Classifier models are trained for each period, and SHAP values are then applied to quantify global and class‑specific feature contributions. Results reveal a dramatic collapse of high occupancy during the economic downturn, a corresponding surge in low occupancy, and partial normalization during recovery. International visitors strongly predict occupancy before and after COVID-19 but lose influence during the economic downturn, when domestic visitors become the primary driver. Employment-related features—especially food and beverage and room staffing—grow in importance during recovery, indicating their role in signaling operational readiness and resilience. Model accuracy peaks in the pre-COVID period (86%) and declines thereafter, reflecting structural instability and evolving demand relationships. Overall, this study demonstrates the value of interpretable AI in capturing crisis-induced structural change in hospitality markets and offers actionable insights for hotel managers and policymakers working to support sustainable recovery through integrated tourism and workforce strategies.