Research And Application Of Second-hand House Price Forecasting Model Based On LSTM

The price of second-hand housing has always been a matter of great concern to people.Its accurate prediction is of great significance to urban planning,public housing purchase,market allocation and supervision.The problem of change,house price prediction has evolved from the mathematical statistics method initially adopted to the deep learning algorithm currently used.In the process of continuously pursuing higher prediction accuracy,the researchers found that the timing of price changes cannot be ignored,and proved that the LSTM neural network model has advantages in dealing with sequence problems.However,in the research of using LSTM model to predict house prices,because the data must be time-series,only the regional average annual price or monthly average price can be predicted.In response to this problem,this paper obtains second-hand housing data in Jinshui District of Zhengzhou City,assigns spatial sequences to the data on the basis of time series,constructs a prediction model based on LSTM neural network,and makes a one-price-one-price prediction.This article makes predictions from the perspective of differences in spatial distribution of data,and has the advantage of being less affected by policies and other factors.The main work of this article is as follows:?In order to obtain more comprehensive second-hand housing information,this article selects the Fangtianxia website for data crawling,adopts the Scrapy-redis distributed crawler framework as a whole,writes crawling rules based on the Fangtianxia website,and collects it at a fixed time every month.Fangtianxia website's second-hand housing data from October 2018 to August 2019.?Through spatial autocorrelation analysis,the spatial association mining of the unit price attributes in the data is based on the spatial correlation between the second-hand housing data.Here,the spatial sequence algorithm based on Haversine is introduced,and the algorithm is used to spatially sequence the monthly data sorting makes each month's data into a long sequence with spatial correlation.?Build a model based on LSTM neural network,design a personalized Attention mechanism layer and improve it according to the sequence characteristics of the data in this article.Without changing the data sequence,the Attention mechanism gives each data an exclusive weight,by highlighting more important features forecast accuracy is improved.This paper designs three experimental schemes based on LSTM neural network for comparison and prediction,using MAE,RMSE and R~2as model evaluation criteria.The experimental results show that the spatial sequence+Attention mechanism+LSTM model constructed in this paper has better prediction effect.