Research On Forecast Of Beijing Second-hand House Transaction Price Based On Stacking Theory

With the rapid development of China's economy and urbanization,China's real estate market is thriving.People pay more and more attention to real estate information.However,there are fewer and fewer land available for development in the city.The second-hand housing market has gradually become an important role in real estate transactions.Accurate demand for second-hand housing valuation is also growing.Quick prediction of housing prices can enable the government to formulate precise regulatory policies and supervise the intermediary market.It can enable intermediaries to keep up with the development trend of the real estate industry and standardize their services.Can allow buyers to use the estimated results as a reference to avoid problems such as intermediary fraudulent buyers.On the real estate price appraisal in China are generally adopt three traditional assessment methods:the market comparison method,cost method and income method.Among them,the market comparison method is the most widely used.However,the use of market method and the experience of the evaluator are closely related,and influenced by subjective factors of the subjective evaluator.In recent years,domestic scholars used statistical modeling to predict house prices in order to improve the shortcomings of market method,and achieved good results.In this paper,Stacking algorithm is introduced to establish a model to predict the house price of second-hand housing.In this paper,the network crawler captures more than 20,000 second-hand house information from the second-hand house website of Beijing Lianjia sites.Its characteristics include 33 features of the second-hand house,such as house type and inner area.Through data exploration,data cleaning,data transformation,data dimension reduction and other processes,an optimal index system is established,and a Stacking theory is constructed.The model of Beijing second-hand housing transaction price forecasting is optimized by adjusting parameters through grid search.Finally,by using the method of five fold cross-validation,SVR algorithm,GBDT regression algorithm,random deep forest regression algorithm,multi-layer perceptron regression algorithm and Stacking method are used to fuse the four algorithms.Finally,through empirical research,it is concluded that Stacking algorithm has the advantages of better stability and less prediction error than single algorithm,and it is worth popularizing and applying in the real estate price prediction industry.