Machine Learning in Hospitality: Interpretable Forecasting of Booking Cancellations

The hospitality industry faces persistent revenue disruption due to booking cancellations, which affect occupancy forecasts, resource allocation, and pricing strategies. This paper presents an interpretable machine learning framework to forecast hotel booking cancellations with high accuracy and transparency. Unlike opaque neural-network approaches, the proposed system employs decision trees, logistic regression, and explainable boosting machines trained on historical booking data encompassing lead time, customer demographics, room category, price levels, and distribution channels. SHapley Additive exPlanations (SHAP) are integrated to quantify individual feature contributions to each prediction, enabling hotel managers to understand cancellation drivers rather than simply receiving a binary outcome. Stratified k-fold cross-validation with random under-sampling addresses class imbalance. Experimental results confirm that interpretable models achieve competitive predictive performance while delivering actionable transparency. Key findings indicate that extended lead time, flexible cancellation policies, and certain distribution channels are the strongest predictors of cancellation. These insights support targeted interventions such as dynamic pricing, personalised promotions, and policy refinement. The study demonstrates that interpretability and predictive accuracy are not mutually exclusive, establishing a practical pathway for AI-assisted decision-making in hospitality management.