Mapping the fine-scale spatial pattern of housing rent in the metropolitan area by using online rental listings and ensemble learning

Abstract

Abstracts The accurate mapping of housing rent is crucial to the understanding of residential dynamics. In this study, we proposed the use of online rental listings as a new reliable data source for mapping housing rent. With the collected individual rental information from an online platform, we attempted to produce the fine-scale spatial pattern of housing rent in the metropolitan area of Guangzhou, China, at the neighborhood committee (NC) level. This involves the task of estimating the housing rent for areas with no observation data of housing rent. To this end, we evaluated six numeric prediction methods of machine learning. We further enhanced their performance through ensemble learning, an approach which can form new classifiers with even better performance than any of the individual constituent classifiers. We implemented ensemble learning through ways of bagging and stacking, and selected the most accurate ensemble classifier to produce the spatial pattern of housing rent at the NC-level. In the resulting housing rent pattern, we identified a distance decay relationship between the housing rent and the distance from the city center. The data sources and the ensemble learning platform in this application of housing rent mapping are generally open access. Therefore, the proposed approach in this study can provide useful hints for housing rent mapping in other geographical areas. Our mapping results can also be integrated with additional information to support the studies of urban residential problems in China.