Study On The Price Evaluation Of Second-hand Housing In Wuhan

As the real estate market becomes more and more perfect,people's demand for real estate information becomes more and more urgent.Fast and accurate real estate valuation can not only provide scientific assessment prices for developers to develop real estate and residents,but also ensure that real estate transaction parties can effectively promote business.China's current valuation of real estate mainly includes three methods:market method,cost method and income method,of which market method is used most frequently.However,the use of market method has its drawbacks in the valuation of real estate.This method relies on the subjective experience and personal qualities of the evaluator.Especially when the number of similar transactions is small,it is prone to large deviations.In recent years,many scholars have used mathematics and statistics knowledge to introduce mathematical models into real estate price assessments.This has made it possible to evaluate properties quickly,in batches,accurately and at low cost.This paper introduces the machine learning algorithm model into the real estate evaluation based on the latest foreign research results and has achieved good results.Based on the data of 2821 sets of second-hand housing in Wuhan,this paper makes statistical analysis in three aspects:First,using Tableau to visualize the data,descriptive statistical analysis of the data,found that factors affecting the second-hand housing price and the conclusion is consistent with subjective cognition.Second,we divide the housing price into different levels,and explore the factors that influence the price level difference.Based on all the sample data,the author established six algorithm models to fit the hierarchical distribution of house prices.According to the model's overall false positive rate and 10-fold cross validation,the model performance was measured and it was found that the random forest performance was optimal.Then,according to the variable importance index given by random forests in classification problem,the factors that influence the price level difference are divided into three categories.Third,the linear regression model,the neural network model and the random forest model were established to evaluate the real estate prices.The author uses a stepwise regression method to optimize the linear regression model;uses the mean square error of the test set to optimize the parameters of the neural network,determines the optimal number of network hidden layers and the number of nodes;and debugs the effect of parameter values in the random forest on the generalization performance of the model,and thus the value of the parameter ntree and mtry were determined.Finally,this paper constructs indicators to evaluate the prediction performance of different models,and gives the average relative error and statistical matching degree interval of the three models,and compares the prediction performance of the model.