Dynamic temporal reinforcement learning and policy-enhanced LSTM for hotel booking cancellation prediction

Abstract

The global tourism industry is expanding rapidly, making effective management of hotel booking cancellations crucial for improving service and efficiency. Existing models, based on static data assumptions and fixed parameters, fail to capture dynamic changes and temporal trends. Real-world cancellation decisions are influenced by factors such as seasonal variations, market demand fluctuations, holidays, and special events, which cause significant changes in cancellation rates. Traditional models struggle to adjust dynamically to these changes. This article proposes a novel approach using deep reinforcement learning techniques for predicting hotel booking cancellations over time. We introduce a framework that combines dynamic temporal reinforcement learning with policy-enhanced LSTM, capturing temporal dynamics and leveraging multi-source information to improve prediction accuracy and stability. Our results show that the proposed model significantly outperforms traditional methods, achieving over 95.9% prediction accuracy, a model stability of 0.98, an F1 Score approaching 1, and a mutual information score of approximately 0.93. These results validate the model’s effectiveness and generalization across diverse data sources. This study provides an innovative and efficient solution for managing hotel booking cancellations, demonstrating the potential of deep reinforcement learning in handling complex prediction tasks.

Keywords

• Reinforcement learning
• Dynamic temporal prediction
• Hotel booking cancellation probability
• Policy enhancement