Hotel booking curves: taxonomy, algorithmic effectiveness and computational efficiency
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Accurate demand forecasting is fundamental for hotel revenue management. This dissertation explores how booking curves—representing cumulative reservations over time—can be better categorized and used to generate more accurate forecasts using k-Nearest Neighbors (k-NN) algorithms. Drawing on over 3.8 million daily booking observations, which were aggregated into more than 70,000 booking curves from 50 U.S. hotels, the research is organized into three studies. ☐ The first study examines whether booking curves can be meaningfully categorized using unsupervised clustering. Both Euclidean-based and adaptive polynomial clustering were conducted for the categorization, it revealed three consistent behavioral segments: early, mid, and last-minute bookers. ☐ The second study investigates how variations in k-NN configuration—such as the number of neighbors (k), distance metrics, and weighting arrangements—affect forecasting accuracy across different booking horizons. Results show that the optimal combination of k-NN features vary by hotel booking horizon length. ☐ The third study evaluates whether pre-clustering booking curves before applying k-NN algorithm improves computational efficiency. Both clustering methods significantly reduced processing time with only modest trade-offs in forecast accuracy. ☐ Overall, these findings offer a practical framework for developing hotel demand forecasting systems that balance accuracy and scalability. By aligning behavioral segmentation with horizon-specific algorithm tuning, this research provides actionable strategies for revenue managers to enhance forecast precision while reducing computational complexity. Additionally, the results demonstrate that algorithm efficiency can be significantly improved through thoughtful clustering, making the approach well-suited for real-time, data-intensive operational environments.
Description
Keywords
Hotel booking curves, Taxonomy, Algorithmic effectiveness, Computational efficiency, k-Nearest Neighbors