Research On Second-Hand Housing Transaction Price Prediction Based On Machine Learning Model

Housing problem is a critical issue related to the people’s livelihood and national development.Pre-owned houses are also an important approach to solve the problem,which is booming nowadays.So the study on the factors affecting the price of pre-owned houses is becoming more and more important.However,most scholars’ research mainly focuses on the listing price.According to the real situation,the listing price is generally higher than the final housing price.Relatively speaking,its reference is not valuable for buyers.Therefore,this paper sets the foothold on the real transaction price of pre-owned houses,which has much more practical significance.Based on the transaction price of pre-owned houses in Hohhot as the research object and the real transaction data Home Link on line to crawl the housing transaction data in Hohhot from October 2018 to December 2021,this paper predicts and analyzes factors influencing on the transaction price of pre-owned houses in the urban area of Hohhot and puts forward corresponding countermeasures as follows.To start with,considering the location,architectural characteristics and transaction features,the initial characteristic variables were established,based on the previous research.Then by selecting representative characteristic variables from the initial ones according to the availability of data,relevant data were obtained through network crawler technology.And the data was transformed to analyze the correlation between feature variables and housing prices.Selected candidates were screened by multiple linear regression and Lasso regression.Finally,there are 35 characteristic variables selected.Besides,the purpose is to further explore the key factors influencing on housing prices,build the prediction model,and select the optimal model through the comparison between model effects.The Random Forest model,XGBoost model and Neural Network model were set up by combining the grid search method in order to compare the average absolute error,average absolute percentage error,root-mean-square error,root-mean-square percentage error,and then determine coefficient evaluation index of model prediction.The result indicates that the XGBoost model is of the best,followed by Random Forest model,and Neural Network model is the worst.The XGBoost model and the random forest model were used to find that the building extent,areas,number of bedrooms,decoration,distance from hospitals and schools have a significant impact on housing prices.What’s more,according to the analysis of the current situation of the pre-owned houses’ market in Hohhot,the main problems in several urban areas of Hohhot are imbalanced urban development,the huge gap of housing price and urgent improvement of market needs,as well as the price asymmetry between buyers and sellers.Therefore,the corresponding suggestions are: Strengthening the regulatory role of the government on pre-owned real estate;promoting the rationalization of sales intermediary pricing and structure standardization;guiding buyers to make appropriate decisions and avoid impulse consumption.