Explainable profit-driven hotel booking cancellation prediction based on heterogeneous stacking-based ensemble classification
提出一种异构线性堆叠集成分类器，通过面向利润的训练和全局事后解释策略，在提高酒店预订取消预测准确性的同时增强模型可解释性，帮助酒店制定更有效的客户保留策略。
The goal of hotel booking cancellation prediction in the hospitality industry is to identify potential cancellations from a large customer base and improve the efficiency of customer retention and capacity management efforts. Whilst prior research has shown that the predictive performance of hotel booking cancellation prediction can be further enhanced by integrating multiple classifiers, the explainability of such models is limited due to low interpretability and limited alignment with company goals. To address this limitation, we propose a novel heterogeneous linear stacking ensemble classifier for profit-driven hotel booking cancellation prediction. It enhances classifier explainability by (1) making models more accountable by axing model training towards profitability and (2) complementing models by global post-hoc model interpretation strategies. Through experiments based on real-world datasets, our proposed classification framework is demonstrated to lead to greater profits than other profit-oriented predictive models. Moreover, an in-depth interpretability analysis demonstrates the framework's ability to identify critical factors significantly impacting hotel cancellations, providing valuable insights for retention campaigns.