| Summary: | In the hotel industry, demand forecast accuracy is highly impacted by booking cancellations. Trying to overcome loss, hotels tend to implement restrictive cancellation policies and employ overbooking tactics which in turn reduces the number of bookings and reduces revenue. To tackle the uncertainty arising from cancellations, models for the prediction of a booking's cancellation were developed. Data from hotels' reservations systems was combined with data from other sources (events, holidays, online prices/inventory, social reputation and weather). Despite data class imbalance, concept drift, and dataset shift problems, it was possible to demonstrate that to predict cancellations of bookings is not only possible but also accurate. Moreover, it helped to better understand what the cancellation drivers can be. In order to assess the models under real conditions, a prototype was developed for field tests allowing to evaluate how an automated machine learning system that predicts booking’s cancellations could be integrated into hotels' systems. The model's performance in a real environment was assessed, including the impact on the business. The prototype implementation enable an understanding of adjustments to be made in the models so that they could effectively work in a real environment, as well as fostered the creation of a new measure of performance evaluation. The prototype enabled hoteliers to act upon identified bookings and effectively decrease cancellations. Moreover, results confirmed that booking cancellation prediction models can improve demand forecast, allowing hoteliers to understand their net demand, i.e., current demand minus predicted cancellations.
|
|---|