Research On Community Average Price Prediction Model Based On Moving Least Square Method

The real estate industry is the cornerstone and driving force for China's economic development.With the increasing activity of real estate transactions,the demand for real estate valuation also increases.In recent years,the mainstream valuation method is the hedonic price method,which takes the price determined by the features of commodities as the basic principle and combines mathematical models to evaluate the price comprehensively.There are a lot of studies on the use of hedonic price method to evaluate residential buildings.There are many features to be selected in the modeling process.It is necessary to collect residential features manually on-the-spot,and the cost is high in practical application.Some researchers put forward that community is the starting point of housing price evaluation.Community is a collection of residential buildings in a certain area.Compared with residential buildings,it has fewer and more macroscopic features and is easy to collect.The overall average price judgment of the community is helpful to fully understand the situation and development trend of housing prices in the region,and reflects the individual price level of housing in the community from the side.At present,there are few studies on the community average price,mainly using the multiple linear regression method commonly used in the hedonic price method to model and analyze data,which relies on the linear assumption of data and the fixed fitting function settings,and is liable to cause larger errors.On the basis of summing up the previous work,this paper introduces the moving least square method to mine and analyze the intrinsic rules of community feature data.The main work is as follows:(1)Construction of community feature data set.Within the theoretical framework of Hedonic Price Method,this paper screens and refines community available features in common residential features,and proposes to use longitude and latitude data instead of traditional location features to reduce feature dimensions.Taking Dongcheng District and Xicheng District of Beijing as the empirical research areas,using the network as the media,the community features and average price data of 2018 in the study area are collected by automatic collection of feature data.Through data integration,cleaning and standardization,10486 7-dimensional data samples of 908 communities are obtained as data sets for modeling and analysis.(2)Research on community average price forecasting model.In this paper,the moving least squares method with high fitting accuracy is introduced to replace the multiple linear regression in hedonic price method for the first time.The basis function and weight function used in the algorithm are discussed and selected.On the basis of the original algorithm,a dynamic influence radius setting method is proposed to solve the problem of low prediction accuracy caused by uneven data distribution in the community average price forecasting scenario.Compared with the model constructed by multivariate linear regression,the results show that the prediction results of the model constructed by moving least squares method have high stability and accuracy,and the model is easy to explain the validity,which is worth popularizing and applying.(3)Design and implement a visual prototype system.According to the completed data sets and models,the prototype system architecture,data flow logic,underlying database design and interface implementation methods are proposed around functional requirements.The functions of cell search,screening,average price prediction and data presentation are realized,and the system functions are tested,providing an intuitive reference for model application from the perspective of software engineering.