Identification And Quantification Of Influencing Factors Of Urban Residential Land Price Based On Deep Learning

With the rapid economic development nationwide in recent years,land transactions in cities have become more and more frequent.At the same time,problems such as land speculation and excessive price increases have also arisen.As an important national asset,land has special importance in the social economy.Therefore,the norms of the land market,the restrictions on land prices,and the control of the real estate industry are major issues that the country needs to address in the future.As a necessity of the people's life,residence has become a top priority in the research of land issues.Due to its importance,the price of residential land is affected by many complex factors.The research of residential land prices is essentially a research of the factors affecting residential land prices.In the present study,although the Hedonic models,Back propagation neural networks and other research methods all have their own advantages,there are certain defects in the face of the complex issue of land prices.Therefore,this paper collected the infrastructure data of Chengdu in 2010 and the residential land transaction data from2010 to 2017.Based on deep learning techniques,the following methods are used to study the influencing factors of urban residential land prices:(1)Use ArcGIS~?,Excel~?and other software to perform spatial analysis,data correction,format adjustment and conversion,and data normalization processing on the data.The data on the characteristics of land prices in Chengdu and the price data of residential land transactions were produced.(2)the correlation analysis function of SPSS~?is used to analyze the correlation between the land price factors and the land transaction price.Based on the analysis results,the index of the factors affecting the land price of Chengdu is determined.On this basis,Hedonic model,Back propagation neural network and Deep Belief Nets model based on land price influence factors are established.(3)In order to verify the effectiveness of the combination of correlation analysis and deep belief network in the recognition and quantification of residential land price factors,the same data are used to test the deep trust network based on the correlation analysis,the uncorrelated analysis network,Hedonic price model and the Back propagation neural network,and the experimental results are compared and analyzed.By comparing and analyzing the experimental results of the Deep Belief Nets and the traditional model,and the experimental results of the Deep Belief Nets in different parameters or experimental data,the following conclusions are obtained:(1)When recognizing and classifying Chengdu's land price factors and residential land price grid values,the average accuracy of the Deep Belief Nets is 84%,the average accuracy of the price regression model is 43.3%,and the average accuracy of the Back propagation neural network is 69.1%.This shows that in the land price evaluation experiment based on the land price influence factors,the Deep Belief Nets has great advantage over the traditional model in the accuracy of the model.(2)When the land price factor data are calculated as neighborhood ten thousand metres or five thousand metres,the Deep Belief Nets has high accuracy in land valuation.At the same time,it is proved that the Deep Belief Nets has some adaptability when the input data scale changes.(3)Through correlation analysis,the Deep Belief Nets converges faster in training than the Deep Belief Nets without correlation analysis.The average accuracy of Deep Belief Nets is 84%,which is 11.8%higher than that of Deep Belief Nets without correlation analysis.It is proved that correlation analysis can improve the training efficiency and accuracy of model.