The Mass Appraisal Study of Taxable Price of Commodity Housing Based on Automated Valuation Models

The purpose of this study is to reveal the construction characteristics and advantages of the random forest automatic assessment model based on the transaction data of new commercial housing in X district of A city, and to introduce a rolling time window to enhance the model performance of random forest and guide its practice in the mass appraisal of taxable price of commodity housing. The methods of this study are GIS spatial information processing, random forest algorithm and rolling time window time series analysis. The research results show that: 1) the number of features in the feature subset at regression tree splitting and the node division rules at regression tree splitting have a large impact on the prediction accuracy of the random forest automatic assessment model. 2) The evaluation accuracy and generalization ability of the random forest automatic evaluation model are significantly better than those of the Hedonic model, with a goodness-of-fit of 98% and an average absolute error percentage of 2.7% in the training and test sets. 3) The introduction of a suitable rolling time window can improve the prediction accuracy and economic explanation ability of the random forest automatic valuation model. In conclusion, the random forest automatic valuation model based on rolling time windows can be widely used for automatic batch valuation of taxable prices of real estate in the study area.