Using Multi-source Data And Machine Learning To Identify The Spatial Pattern Of Vacant Housing

Housing vacancy rate is an important indicator of the real estate market bubble,the extent of urban development and resource allocation,and an important"Monitors"of the rational development of the real estate industry,which also directly affects residents’living standards and quality of life.Long-term rapid economic growth and the prosperity of the real estate industry in China have obscured the phenomenon of vacant houses.Because of the opacity of real estate data,academics cannot agree on the distribution of vacant houses,housing vacancy rates,and other basic data,let alone understand their spatial patterns and dynamic changes.A theoretical framework for creating a scientific urban development strategy as well as a foundation and point of reference for achieving sustainable urban development will be provided by using a rational method to calculate the housing vacancy rate scientifically and analyze the spatial distribution pattern of housing vacancy and its influential mechanism.In order to predict the housing vacancy rate in the Anning District of Lanzhou City in 2022,this study employs remote sensing image data,geospatial big data,socioeconomic statistics,and other data,as well as a variety of machine learning techniques and GIS platforms.In Anning District,Lanzhou City,the spatial distribution pattern of housing vacancy was examined based on multi-source data,and the extent to which each factor affected housing vacancy was investigated by examining the correlation between population distribution,accessibility,house price rent,and vegetation cover and housing vacancy rate using GIS spatial analysis tools.The study’s findings following three conclusions.(1)The BP neural network has the highest accuracy when estimating the housing vacancy rate of Anning District.The BP neural network performs better than the other five types of machine learning models in terms of estimation accuracy and processing speed when calculating the housing vacancy rate.The R~2 of BP neural network is 0.91,much better compared to the other machine learning techniques,and its RMSE is 0.006,indicating that its prediction outcomes are the most accurate.Also,compared to other machine learning techniques,the MSE and bias are reduced.As a result,when it comes to prediction accuracy,BP neural network,XGBoost,and multiple LR have the best prediction effects,whereas RF,SVM,and KNN models have worse prediction effects and more variable results.(2)In Anning District,vacant housing is spatially concentrated overall and dispersed in localized pockets.The overall spatial distribution pattern of housing vacancy rate reveals a low,congested,and continuous housing vacancy rate in the urban center and a high,fragmented,and discrete housing vacancy rate in some areas surrounding the urban area and on Shajingyi Street.These patterns tend to be consistent with the spatial distribution patterns of night lighting and POI.Using the BP neural network,the average housing vacancy rate in Lanzhou City’s Anning District was predicted to be 15.04%,with housing vacancy rate of less than 20%in 58.4%of the areas.The overall vacant areas are contiguous,with some areas having spatially discrete housing vacancies and wide variations in housing vacancy rates between adjacent grids.(3)There is a significant positive correlation between housing price and rent and a significant negative correlation between housing vacancy rate and population and vegetation cover,there is spatial consistency between housing vacancy rate and the distribution of roads and bus stops.Following analysis,it was discovered that the housing vacancy rate and population had a negative spatial correlation in 79.51%of the areas,with a correlation coefficient of-0.68.Low housing vacancy rate areas were mostly found in easily accessible locations,like public transportation hubs.housing vacancy rate and vegetation cover have a negative spatial correlation as well.Notably,there is a noticeable spatial correlation between the locations of low housing vacancy rates and the locations of high vegetation cover in the periphery.