ERROR BEHAVIOR ANALYSIS IN DATA-DRIVEN MODELS FOR ESTIMATING HOTEL DEMAND DYNAMICS

Abstract

Rapid and irregular changes in room demand patterns pose a major challenge for Hotel U in ensuring effective operational planning and revenue strategies. This instability is influenced by the dynamics of last-minute bookings, seasonal variability, and competition on digital platforms. To address these conditions, this study evaluates two data-based modeling approaches, namely Linear Regression and K-Nearest Neighbors (KNN), with the aim of assessing the accuracy of revenue estimates and the consistency of the two models' performance on different proportions of training data. Three data distribution schemes were used 70:30, 80:20, and 90:10 to observe the sensitivity of the models to changes in the amount of historical information. The results show that KNN is more effective on data with high diversity, marked by an RMSE value of 0.057 in the 70:30 scheme. These findings indicate that model selection should be tailored to the analysis objectives: neighbor-based models are superior for short-term changes, while linear models are more appropriate for income movement patterns that follow general trends.