A comprehensive approach to enhancing short-term hotel cancellation forecasts through dynamic machine learning models

Enhancing the accuracy of short-term forecasts for cancellation rates offers revenue managers the opportunity to formulate a pricing strategy for the upcoming day, yielding favorable economic outcomes. This study proposes a methodology based on dynamic models utilizing machine learning methods such as LightGBM, XGBoost, Random Forest, and ANN-MLP, highlighting the importance of data dimensionality reduction while using higher-performing variables to improve predictive accuracy and stability. Notably, lagged variables within a few days of the forecasted date and reservations made through OTAs and BAR-rated reservations exhibit significant predictive power. The results indicate that artificial neural network multi-layer perceptron (ANN-MLP) outperforms other models, especially in longer forecast horizons. The study recommends adaptable strategies considering historical data and temporal trends and leveraging ANN-MLP for superior accuracy. The findings offer valuable insights for industry practitioners, providing a nuanced understanding of cancellation patterns and suggesting strategies to optimize cancellation prediction models in a competitive marketplace.