Predicting hotel bookings cancellation during high volatility times: Using machine learning for hotel booking management during a pandemic

Dissertation presented as the partial requirement for obtaining a Master's degree in Information Management, specialization in Knowledge Management and Business IntelligenceLike many other service industries, booking cancellations highly impact hotel management decisions, negatively contributing to accurate forecasts. With accurate forecasts, hotels can make better decisions about bookings to accept and reject, what prices to make, and how many rooms to oversell, leading to higher revenue and lowering costs. Previous research showed it is possible to develop good-performing predictive models using hotel booking data. However, past research was not developed considering a pandemic where mass cancellations happen. Thus, it was not yet verified if performance is affected by those new circumstances. In a first study, using hotel booking data from four hotels, this new research assesses the machine learning classification models’ performance under the conditions imposed by the COVID-19 pandemic, achieving fair to excellent performance. With city hotel C1 achieving 0.72 AUC, 0.80 Accuracy, and 0.76 F1-Score, C2 achieving a 0.94 AUC, 0.93 Accuracy, and 0.94 F1-Score. Resort hotel R1 achieved 0.72 AUC, 0.59 Accuracy, and 0.69 F1-Score, and lastly, R2 achieved 0.70 AUC, 0.56 Accuracy, and 0.64 F1-Score. Besides the predictive performance, this study highlights the critical predictive features. The lead time was the most crucial feature of the XGBoost model for three of the four hotels. A second study implemented a sliding window technique to understand how the performance differs in different timespans. For the configuration of twenty-four months of training and nine months of test data, it was possible to achieve better results than in the previous study. Two windows were created, with the second window having nine months of training data since the pandemic's start. For the first window the AUC varied from 0.80 to 0.98 on test data. For the second window the AUC varied from 0.85 to 0.99 on test data. This research highlights the importance of machine learning in hospitality management, particularly during a crisis such as an unprecedented pandemic. Both studies show that it is possible to create machine learning classification models to predict hotel booking cancellations during a pandemic with fair to excellent results. Results of the second study emphasized the importance of having newer data in model training and constant monitoring. Artificial Intelligence models can help managers improve their results by providing superior accuracy in a timelier way