Machine Learning-Driven Hotel User Behavior Analytics and Intelligent Decision Support: A State-of- the-Art Review

Abstract

Machine-learning studies in hotel analytics do not address one common unit of analysis. Some predict an event at booking level, some interpret language at review level, and others estimate value at customer level. This review examines how those separate tasks may contribute to hotel decisions. The papers considered cover order propensity, guest-experience mining, cancellation risk, customer lifetime value, mobile reservation behavior, and explainable modeling. Their samples, targets, and evaluation measures differ, so the review does not rank algorithms by reported accuracy. Instead, it asks what evidence each model produces, who could use that evidence, and what remains to be decided after a prediction is made. Read together, the studies point to a five-part process: customer records are assembled, relevant features are constructed, a task-specific estimate is produced, the estimate is interpreted, and a manager selects an action. This sequence also exposes gaps in the literature. Structured records and review text are not routinely connected; explanations are often technical rather than operational; changing market conditions receive less attention than model fit; and the effect of the eventual hotel intervention is seldom measured. A closed-loop framework is proposed to organize these gaps. The framework treats a later customer response as evidence for revising both the model and the decision rule. Machine learning, on this account, informs hotel judgment but does not replace it.