Predicting Cancellations of Hotel Reservations with Unsupervised Machine Learning Methods

This research discovers explores the applications of clustering and anomaly detection techniques to recognize identify high-risk hotel bookings prone to cancellation. The study utilizes K-Means and DBSCAN for unsupervised clustering, along with the Isolation Forest algorithm for detecting anomalies. K-Means clustering provides an initial segmentation of bookings, highlighting groups with higher cancellation rates, but its limitations in handling noise led to the implementation of DBSCAN. DBSCAN identifies clusters based on data density and labels outliers, offering a more flexible approach that effectively distinguishes typical bookings from anomalies. Isolation Forest further validates the findings by identifying unusual bookings, overlapping with DBSCAN's noise points, and providing consistent insights. The results show that DBSCAN and Isolation Forest are effective in identifying high-risk reservations, providing hotels with actionable insights to minimize cancellations. The study concludes that using density-based clustering and anomaly detection allows for a more refined understanding of booking behaviors, leading to better-targeted strategies for revenue optimization.