Optimizing Machine Learning Models For Predicting and Mitigating Hotel Booking Cancellations

Hotel booking cancellations pose substantial challenges to the hospitality industry, significantly impacting revenue management and operational planning. This study explores the application of machine learning models to predict cancellations, emphasizing model selection, feature importance, and resampling techniques. Among the six classification models evaluated, the combination of XGBoost and SMOTE demonstrated the highest predictive accuracy and consistency. Feature importance analysis and SHAP interpretation identified key predictors, including deposit type (non-refundable), required parking spaces, previous cancellations, and market segment (OTA). Additionally, threshold tuning was examined to balance the trade-off between false positives and false negatives based on business priorities. The results underscore the critical role of resampling methods in addressing class imbalance and the necessity of optimizing classification thresholds for practical deployment. Future research will focus on advanced hyperparameter tuning, alternative resampling strategies, feature selection methods, and ensemble learning approaches to enhance model robustness and interpretability. These findings provide a data-driven foundation for improving cancellation prediction and guiding strategic decision-making in hotel management.