BOOKER PREDICTION FROM REQUESTS FOR QUOTATION VIA MACHINE LEARNING TECHNIQUES

Abstract

Purpose – Many incoming requests for quotation usually compete for the attention of accommodation service provider staff on a daily basis, while some of them might deserve more priority than others. Design – This research is therefore based on the correspondence history of a large booking management system that examines the features of quotation requests from aspiring guests in order to learn and predict their actual booking behavior. Approach – In particular, we investigate the effectiveness of various machine learning techniques for predicting whether a request will turn into a booking by using features such as the length of stay, the number and type of guests, and their country of origin. Furthermore, a deeper analysis of the features involved is performed to quantify their impact on the prediction task. Findings – We based our experimental evaluation on a large dataset of correspondence data collected from 2014 to 2019 from a 4-star hotel in the South Tyrol region of Italy. Numerical experiments were conducted to compare the performance of different classification models against the dataset. The results show a potential business advantage in prioritizing requests for proposals based on our approach. Moreover, it becomes clear that it is necessary to solve the class imbalance problem and develop a proper understanding of the domain-specific features to achieve higher precision/recall for the booking class. The investigation on feature importance also exhibits a ranking of informative features, such as the duration of the stay, the number of days prior to the request, and the source/country of the request, for making accurate booking predictions. Originality of the research – To the best of our knowledge, this is one of the first attempts to apply and systematically harness machine learning techniques to request for quotation data in order to predict whether the request will end up in a booking.