Enhanced Predictive Analytics for Hotel Booking Cancellations: A Fusion Approach Integrating Artificial Neural Networks and Stacking Techniques

This paper introduces an advanced predictive analytics model for hotel booking cancellations, integrating Artificial Neural Networks (ANN) optimized by Genetic Algorithms (GA), Support Vector Machines (SVM), and Random Forest (RF) within a stacking ensemble framework. The study aims to mitigate significant revenue loss caused by cancellations in the hotel industry through the application of machine learning techniques that accurately predict booking cancellations. Utilizing a dataset comprising nearly two years of real-world data from high-star hotels in Guizhou Province, China, which includes 17 features and 102,897 observations, the proposed model leverages the strengths of SVM for handling high-dimensional data, RF for robust feature selection and ensemble learning, and GA-optimized ANN for capturing complex non-linear relationships. Results indicate that the stacking ensemble model outperforms individual models, achieving superior accuracy, precision, F1-score, and AUC. Furthermore, a model with reduced feature sets maintains high performance metrics, offering practical advantages in computational efficiency and resource requirements. The study concludes that the integrated model provides a robust solution for predicting hotel booking cancellations, thereby enabling hotels to optimize their decision-making processes and enhance operational efficiency and revenue management. Additionally, the research underscores the importance of customer historical data for management decisions in the hospitality industry and suggests future research directions, including the exploration of diverse datasets and the incorporation of additional variables to further improve prediction accuracy.