Identifying urban households in relative poverty with multi-source data: A case study in Zhengzhou

Most existing poverty research focused on the identification of households in absolute poverty; few studies attempted to identify households in relative poverty (HRP). In this paper, we first developed an improved Bayes's theorem algorithm to sense individuals' spatio-temporal behavior characteristics via integrating mobile phone signaling, electricity consumption, and Points of Interest. We then utilized the improved random forest to identify HRP based on individuals' spatio-temporal behavior characteristics and building properties. A total of 29,370 urban households in Fengchan community, Zhengzhou, China, were selected to conduct this study. The accuracy rate was about 90% when it was verified against the household survey data. Three conclusions can be drawn from our analysis: (1) the individuals' spatio-temporal behavior characteristics played a more critical role in identifying HRP than building properties, (2) the identification accuracy of multi-source data is higher than that of single-source data, (3) mobile phone signaling records and building footprints data are more important in identifying HRP in low-rise buildings, while electricity consumption data is more crucial in the identification in high-rise buildings. Our proposed methods can accurately identify urban HRP, which is helpful to target interventions in the most needed areas. Our findings can inform relief policies in similar cities.