Interpretable machine learning for hotel demand prediction: A case study framework from Xiamen, China

The advent of big data has led to increased recognition of the benefits of machine learning models in the field of demand prediction. However, the limited transparency and interpretability of most existing machine learning models raises concerns regarding their applicability and effectiveness in facilitating informed decision making. Based on Subjective Expected Utility Theory, the present study attempts to address this challenge by proposing an interpretable machine learning framework, which contains a prediction module and an interpretation module. Results show that the framework enables accurate predictions and offers explanations of the predictions from the following four perspectives: feature importance, feature contribution, nonlinear relationship, and feature interaction. This study validated the prediction accuracy and interpretability of our framework through an ex-post analysis of demand data from two hotels in Xiamen, China. This study contributes to the hotel demand prediction literature and improves management decision making and resource optimization.