Monitoring housing rental prices based on social media:An integrated approach of machine-learning algorithms and hedonic modeling to inform equitable housing policies. (March 2019)

• Record Type: Journal Article
• Authors: Hu, Lirong; He, Shenjing; Han, Zixuan; Xiao, He; Su, Shiliang; Weng, Min; Cai, Zhongliang
• Is Part Of: Land use policy. Volume 82(2019)
• Page Start: 657
• Page End: 673
• Publication Date: 2019-03
• Subjects: Housing rental price -- Social media -- Hedonic model -- Multilevel determinants -- Machine learning -- Equitable housing policies; Land use -- Periodicals; Land use -- Government policy -- Periodicals; Sol, Utilisation du -- Périodiques; Sol, Utilisation du -- Politique gouvernementale -- Périodiques; Electronic journals; 333.7305
• DOI: 10.1016/j.landusepol.2018.12.030
• Languages: English
• ISSNs: 0264-8377

Abstract

Abstract: National land use policies and strategies worldwide have attempted to establish a healthy housing rental market towards urban sustainability. Monitoring fine-scale housing rental prices should provide essential implications for equitable housing policies. However, doing so remains a challenge because aggregated data were traditionally collected at a coarse scale through census or social surveys. On-line housing rental websites (OHRWs) have become popular social media platforms in the housing studies. This paper attempts to demonstrate how to monitor fine-scale housing rental prices based on OHRWs using the case of Shenzhen in China. Employing hedonic model, a set of housing rental determinants are initially selected from three characteristics (neighborhood, location and structure) and at three levels (nearest accessibility, 15-minute walking distance availability and sub-district availability). Housing rent prediction models are then established (respectively for October 2017 and February 2018) using the training samples collected from the OHRWs and six machine-learning algorithms, including random forest regression (RFR), extra-trees regression (ETR), gradient-boosting regression (GBR), support vector regression (SVR), multi-layer perceptron neural network (MLP-NN) and k -nearest neighbor algorithm ( k -NN). Thereafter, the relative importance of the determinants is calculated and visualized using partial dependence plots. Finally, the models are used to monitor housing rental price dynamics for all of the communities within Shenzhen. Results show that all of the algorithms except SVR generally present good performance. Among them, RFR and ETR are the best one in October 2017 and February 2018, respectively. Concerning the spatial pattern of housing rental, the high-high clusters merge in the central districts, whereas the low-low clusters are located in the outskirts, and the growth rate is the greatest in the farthest outskirts from the central districts. Each determinant affects the housing rent across different scale and sub-district availability and nearest accessibility are more important than 15-minute walking distance availability. The two most influential determinants are sub-district job opportunity and nearest accessibility to health care facilities. The case of Shenzhen shows that the demonstrated framework, which integrates machine-learning algorithms and the hedonic modeling, is practical and efficient. The approach is believed to provide an essential tool to inform equitable housing policies.