Research On Mixing Multi-factors Real Estate Price Forecasting Model Based On Investors' Attention

The penetration of Internet technology in people's daily life has changed the operation mode of the real estate market,expanded the ways and channels for investors to obtain information.With the development of behavioral economics,more and more scholars began to research invertors' behavior,meanwhile,the behavioral elements of investors are recorded in the Internet data.Under the premise of protecting user privacy,we could use Internet-related data to describe investors' behavior better and support the project of forecasting real estate price.The research based on search behavior for investors' attention has yielded fruitful results.However,internet data contains a large number of unstructured data,and due to collation,publication and some other reasons,there are also inconsistencies in the frequency of time statistics.In addition,with the development of intelligent technology,more intelligent learning methods were applied to the real estate price forecasting problem.Under this background,based on the complexity and non-linearity of the real estate market,this paper introduces text data and extracts the key words used to measure the investors' attention through text mining.At the same time,in order to persist the original information of mixing data and pursue higher reliability of prediction results,a mixed-frequency multi-factor real estate price forecasting model based on support vector machine and mixing data sampling model is constructed in this paper.At first,this paper describes the research background,significance and innovation of the research subject.Secondly,this paper summarizes and analyzes real estate price forecasting methods and models,the mixed data processing methods involved,and investor-related research at home and abroad,after this,this paper illustrates the feasibility and effectiveness of using Baidu index as the agent variable.Simultaneously,from the qualitative point of view,it analyzes the intrinsic relationship between investors' concern and real estate price fluctuations.Through the collation and analysis of the common real estate price forecasting methods,considering the scope of application and assumption of these models,support vector machine and mixing data sampling model are selected as foundation models.This paper describes the theoretical system,advantages and disadvantages of the support vector machine and the mixed data sampling model.Combining the advantages of both models,a new mixed multi-factor real estate price forecasting model capable of processing nonlinear and mixed data is constructed,give the process of soling model,which expands application scope compared with the basic model,the mixed data exists not only between the explanatory variables and interpreted variables,but also among the explanatory variables.Simultaneously,in order to solve the problem that the kernel function and related parameters in support vector machine are difficult to determine,a combined positive definite kernel function is constructed,and the particle swarm optimization algorithm is used to weight and parameter optimization,which further improves the applicability to different types of data sets.This paper automatically obtain relevant Internet data based on crawler scripts.In order to determine the Baidu Index Keywords,this paper collect text messages from WeChat which is very hot in the mobile Internet era,then using mining technology and random forest encapsulation feature screening to obtain the keyword combination,which realize the conversion of unstructured data to structured.Considering the commonly used macroeconomic and industry-related indicators,a multi-factor impact indicator system is constructed.Six domestic first-and second-tier cities: Beijing,Shanghai,Guangzhou,Shenzhen,Chongqing,Tianjin and Nanjing were selected as research objects to predict the sales price index of new commercial residential buildings,and several experiments are designed.The results show that the introduction of Baidu Index data as investors' attention agent variables can enhance the credibility of real estate price forecast result to a certain extent;simultaneously,the mixed multi-factor real estate price forecasting model constructed in this paper can improve the accuracy and stability of the prediction results compared to co-frequency prediction models,and the use of combined kernel function can further improve the predictive ability of the model.