Network space analysis–based identification of road traffic accident hotspots: a case study

Abstract

This study proposed a method for identifying traffic accident (TA) hotspots. The method combines kernel density estimation (KDE) with network kernel density estimation (NKDE). The hotspots identified through NKDE can be overlaid on the high-risk areas (hot zones) identified using planar KDE, resulting in accurate hotspot identification. The research site was Zhenjiang City in Jiangsu Province, China; data on 410 fatal traffic accidents from 2017 to 2020 were collected. An average nearest neighbour (ANN) ratio (0.563) was obtained using the global auto nearest neighbour distance method; thus, the fatal TAs had a cluster-type distribution in the city. Subsequently, the maximum clustering distance (7812.842591 m) in the research site was analysed using Ripley’s K-function, which yielded a reasonable KDE bandwidth and enabled identifying the distribution of hot zones for fatal TAs. Precise hotspot identification was achieved by overlaying the results of NKDE-based analysis in these hot zones. The proposed method was verified using data on 131 fatal TAs during January to July 2021. The results revealed that, in Zhenjiang City, the identification rate for hot zones and hotspots in hot zones was 71.75% and 38.29%, respectively, and the overall hotspot identification rate was 27.48%, demonstrating the method’s ability to relatively accurately identify hotspot locations, information that can help traffic authorities implement preventive measures.