Traffic crash analysis with point-of-interest spatial clustering

Abstract

This paper presents a spatial clustering method for macro-level traffic crash analysis based on open source point-of-interest (POI) data. Traffic crashes are discrete and non-negative events for short-time evaluation but can be spatially correlated with long-term macro-level estimation. Thus, the method requires the evaluation of parameters that reflect spatial properties and correlation to identify the distribution of traffic crash frequency. A POI database from an open source website is used to describe the specific land use factors which spatially correlate to macro level traffic crash distribution. This paper proposes a method using kernel density estimation (KDE) with spatial clustering to evaluate POI data for land use features and estimates a simple regression model and two spatial regression models for Suzhou Industrial Park (SIP), China. The performance of spatial regression models proves that the spatial clustering method can explain the macro distribution of traffic crashes effectively using POI data. The results show that residential density, and bank and hospital POIs have significant positive impacts on traffic crashes, whereas, stores, restaurants, and entertainment venues are found to be irrelevant for traffic crashes, which indicate densely populated areas for public services may enhance traffic risks.

Keywords: Kernel density estimation; Spatial clustering; Spatial regression; Traffic crashes.