Comparing effectiveness of point of interest data and land use data in theft crime modelling: A case study in Beijing

Abstract

To promote the healthy development of cities, previous studies have long investigated the relationships between urban functions and crime. However, the use of either land use data or point of interest (POI) data to represent urban functions can yield inconsistent findings, potentially misguiding urban planners in crime prevention efforts. To address this issue, we systematically compare the effectiveness of land use and POI data in theft crime modeling with a case study of Beijing, China. Urban function features are constructed from both data sources by three measures, i.e., density, fraction, and diversity. Their global strengths are evaluated through negative binomial regression (NBR). Additionally, geographically weighted negative binomial regression (GWNBR) is employed to uncover their local strengths. Results indicate that POI data generally outperform land use data, with POI densities being the most effective. Nevertheless, optimal data sources and measures vary for urban functions and spatial context. Land use fractions could effectively capture large-scale functional areas, while POI fractions and POI densities are fit for small-scale facilities with distinct properties. This study advocates the complementary use of land use and POI data, offering valuable insights for urban planners and researchers to construct precise urban function indicators for crime modeling.