A Comprehensive Framework to Identify and Classify Traffic Accident Hotspots and Detect Contributing Risk Factors to the Formation of Hotspots

Identifying roads’ hazardous locations and solving their problems are the key measures in traffic safety management. However, since the traditional hotspot identification (HSID) rests on the yearly-aggregated crashes, two problems appear: the locations that become unsafe at specific short periods may remain unidentified as they may not show noticeable crash counts, and the results of the problem diagnosis analysis on hotspots’ crashes potentially contain a great amount of uncertainty. Even though researchers have recently added the dimension of time and analyzed accidents spatio-temporally to obtain more insights, the mentioned problems have not been addressed fully. Hence, this study first suggests a new linear DBSCAN-based HSID method and demonstrates its acceptable performance by comparison with KDE+, the well-known clustering technique; second, employing the proposed technique, the study presents an algorithm for the spatial analysis of accidents through diverse time dimensions, which categorizes the risky locations based on their periodic reappearance. The tempocategorization purpose is to enhance diagnosing causative risks by understanding their arising periods. The algorithm is tested using Allegheny highways crash data from 2014 to 2019. Results illustrate the contribution of the suggested method to the problem diagnosis and for detecting hidden unsafe points. Keywords: traffic accidents, hotspot identification, DBSCAN clustering, spatiotemporal analysis, KDE+, safety problem diagnosis.