A carregar...
Publicação
Timing is Everything: Predicting Hotel Booking Cancellations with Survival Analysis
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.4 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Booking cancellations affect both revenue and resource allocation in the hospitality industry.
To mitigate the risks associated with cancellations, hotels implement a combination of
overbooking and cancellation policies. However, both negatively impact the hotel’s revenue
and reputation. Therefore, it becomes crucial to anticipate cancellations and take proactive
measures to prevent them. While previous research has focused primarily on predicting
whether a booking will be canceled, this study investigates when cancellations are most likely
to occur. As poorly timed interventions can lead to inefficiency and ineffectiveness when
targeting cancellations, it is necessary to predict when cancellations are likely to occur. To
address this, this study analyzes the likelihood of a booking remaining active from the time it
is made until arrival using Survival Analysis (SA), which is traditionally applied in medicine to
estimate a patient's probability of survival over time. This study focused on bookings made at
least 30 days before arrival. Data from four hotels, two city and two resort hotels, were
analyzed using three SA models: Gradient Boosting Survival Analysis, Random Survival Forest
(RSF), and Extra Survival Trees. RSF produced the strongest performance across the assessed
evaluation metrics (Concordance Index, Integrated Brier Score (IBS), and Integrated Area
Under the Curve), particularly for one hotel, where the model achieved an IBS value of 0.04.
By predicting the cancellation day, we accurately identified the exact cancellation day for 69%
of canceled bookings at the same hotel. Across all hotels, when applying time windows around
the predicted day, ranging from 1% to 4% of the lead time, the actual cancellation day fell
within the window in 33% to 72% of cases. Survival probabilities on the arrival day were also
analyzed as a complementary indicator for identifying bookings that were likely to be
canceled. This is the first study to apply SA to hotel booking cancellations, estimating when
cancellations are most likely to occur and introducing a new binary perspective on booking
outcomes. This study provides hotels with a practical tool to support proactive customer
engagement, reduce cancellations, and enhance operational planning and resource allocation.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Business Analytics
Palavras-chave
Booking Cancellation Prediction Hospitality Machine Learning Revenue Management Survival Analysis SDG 8 - Decent work and economic growth SDG 9 - Industry, innovation and infrastructure