Modeling fine-scale residential land price distribution: An experimental study using open data and machine learning

Abstract

Modeling the fine-scale spatiotemporal distribution of residential land prices (RLPs) is the basis for scientifically allocating land resources, managing the residential market and improving urban planning. The accurate mapping of the RLP dynamics require reliable land price prediction models and data with fine spatial and temporal resolution. With the aid of point of interest (POI) data and nighttime light (NTL) images, this paper attempts to explore the ability of machine learning algorithms (MLAs) to model grid-level RLPs using the case of Wuhan in China. Several land price prediction models were built using five MLAs and various geographic variables. The experimental results show that the extra-trees regression algorithm and the radial basis function-based support vector regression algorithm perform best in Period Ⅰ (2010–2014) and Period Ⅱ (2015–2019), respectively; therefore, they were selected to estimate the RLPs of the grids without observations in the corresponding period. Based on the estimated results, we found that the spatial pattern of the RLP in Wuhan transitioned from monocentric to polycentric between the two periods, and RLPs grew rapidly near newly formed urban subcenters and waterscapes. The relative importance of the predictor variables shows that commercial and educational facilities are important determinants of the RLP distribution in Wuhan; moreover, the relative importance of natural amenities and education facilities increased over time, while that of commercial facilities and public transportation decreased slightly. The case of Wuhan confirms the feasibility of MLAs and openly accessible urban data in modeling fine-scale RLP distributions. Our proposed framework provides a new approach to monitor the urban land price dynamics accurately and closely, which is beneficial for improving the infrastructure layout and achieve smart city growth.