Apartment rent prediction using neighborhood rent information based on machine learning and geospatial information

Abstract

The outstanding balance of apartment loans over the medium to long term has been increasing at financial institutions. Financial institutions are faced with the challenge of properly assessing the profitability of apartments and managing the risk. In assessing profitability, it is necessary to accurately predict apartment rents. In this study, we show that the accuracy of rent prediction is improved by using latitude and longitude information and address information, and by using past rents of similar properties in the neighborhood with machine learning model. In addition, we use latitude and longitude information to incorporate geospatial information system(GIS) data, such as information on disasters to which each property belongs, and verify whether this contributes to improving the accuracy of rent predictions. Including the variables of past rents in the neighborhood to the model, we show that information on disasters do not improve the accuracy of rent prediction.