Research on Stock Housing Price Prediction Based on XGR-Stacking Model

• 1、Nanjing Tech University School of economics and management,Nanjing 211800

Abstract：China has now entered an era of stock housing. With the transaction prices of stock housing in cities like Shanghai, Nanjing, and Hangzhou being hidden, market information asymmetry has increased, complicating and making transactions more difficult. Additionally, the "old-for-new" policy in the real estate industry encourages residents to sell stock housing and purchase new homes. Therefore, accurately predicting stock housing prices is particularly important. Currently, machine learning is widely used for housing price prediction, but most models used are single models, which tend to suffer from overfitting and limited expression capabilities. To overcome these shortcomings, this study uses transaction data of stock housing in Nanjing. Through correlation analysis, the 12 factors most influencing transaction prices were selected to establish three single models: XGBoost, Random Forest, and GBDT, along with the XGR-Stacking ensemble model. Using Mean Squared Error (MSE) and the coefficient of determination (R2) as evaluation criteria, results show that the XGR-Stacking model has an R2 of 0.88 and an MSE of 1872, both superior to the single models. This demonstrates that the XGR-Stacking ensemble model has significant advantages in housing price prediction.

Keywords： Housing Price Prediction；Machine Learning；GBDT；Stacking；Bayesian optimization；XGBoost