Forecasting Airbnb prices through machine learning

Abstract

Achieving accurate pricing is critical for both peer-to-peer (P2P) accommodation platforms and hosts. An understanding of the determinants of prices on P2P platforms, such as Airbnb, can improve service quality and help make pricing more rational. In this study, machine learning (ML) was applied to P2P accommodation pricing prediction. Data from Airbnb listings in Sydney, Australia, was collected, and 10 ML algorithms were used to predict prices. Host data were divided into training and testing sets. A total of 35 variables, including price and 34 independent variables, were identified. The 10 algorithms were evaluated using the Student's t test, the root mean squared error, and the R2 value. The CatBoostRegressor algorithm had the best performance. According to the relative weights in the optimized CatBoostRegressor algorithm, the key factors affecting pricing are the maximum number of guests, the number of bedrooms, and whether the room is private. Platforms can use these results to share accurate rental pricing information with hosts. Registered hosts can obtain timely information regarding the house rental market to set reasonable prices.