Research On Valuation Model Of Beijing Second-hand Housing Based On Random Forest

With the development of real estate market, people’s living standard is raising. The real estate information is needed for both the Developers and residents. Therefore, there is an urgent need to establish appraisal system of the real estate for both sides of the deal and the national macro-control. Scientific and fair valuation of real estate, not only provides a scientific measure of value for real estate transactions, but also promotes the transaction services of the real estate. This system is conducive to safeguarding the safety of transactions, as well as the market order.Market comparison approach, cost approach and income approach are the most commonly used methods for the appraisal of the real estate, especially market comparison approach. But this method mainly depends on the experience of appraisers. With insufficient comparable cases, big errors could be caused. Random Forest (RF) is a new machine learning method based on the statistical learning theory. RF can be adopted to make regression and classification analysis via sampling the data continually. It solves the fitting of the nonlinearity and limited data, and have better generalization and the steady frame. Thus, RF can be utilized in various areas.This article puts forward the application of RF in the appraisal of the real estate. It introduces the concept, theory and method of RF, and sets the criterion of the scalar based on Hedonic pricing model. This article takes Beijing Second-hand housing market for example, selects the parameters through trial and error, including the value of mtry and ntree, and applies the optimal parameters to this model. For testing the accuracy and stability of the model, it used the multiple linear regression, CART, Bagging, neural network, SVR to make a comparison with the RF model. The evaluation results show that the second-hand housing pricing evaluation model based on Random Forest performs better in accuracy and stability. This workable evaluation model should be promoted widely.