Spatial Prediction of Housing Prices in Beijing Using Machine Learning Algorithms

Abstract

The real estate industry places key influence on almost every aspect of social economy given its great financing capacity and prolonged upstream and downstream industry chain. Therefore, predicting housing prices is regarded as an emerging topic in the recent decades. Hedonic Regression and Machine Learning Algorithms are two main methods in this field. This study aims to explore the important explanatory features and determine an accurate mechanism to implement spatial prediction of housing prices in Beijing by incorporating a list of machine learning techniques, including XGBoost, linear regression, Random Forest Regression, Ridge and Lasso Model, bagging and boosting, based on the housing price and features data in Beijing, China. Our result shows that compared to traditional hedonic method, machine learning methods demonstrate significant improvements on the accuracy of estimation despite that they are more time-costly. Moreover, it is found that XGBoost is the most accurate model in explaining and prediciting the spatial dynamics of housing prices in Beijing.