A two-phase clustering approach for traffic accident black spots identification: integrated GIS-based processing and HDBSCAN model

Abstract

Identifying black spots effectively and accurately is a pivotal and challenging task to improve road traffic safety. A novel black spot identification model is proposed by integrating the GIS-based processing with hierarchical density-based spatial clustering of applications with noise. Additionally, the optimal clustering parameters are determined based on an internal validation indicator called the density-based clustering validation index to minimize the impact of subjectivity in parameter selection. The model is validated by collecting 3536 accident data from 1 August to 31 October 2020 in Hangzhou, China, and eventually identifies 39 black spots. The results show that: (1) The number of accidents contained in black spots account for 75% of all accidents, while the length of network in the black spots only account for 23.26% of the total road network length. (2) Compared with the conventional density-based spatial clustering of applications with noise model and K-means model, the proposed model achieves the best performance with more accidents gathered per unit road length. (3) The sample survey with 6 onsite of the identified black spots indicates that the proposed model has high recognition accuracy and recommend these sites for further investigation.