Dijkstra’s-DBSCAN: Fast, Accurate, and Routable Density Based Clustering of Traffic Incidents on Large Road Network

Abstract

Incident hotspots are used as a direct indicator of the needs for road maintenance and infrastructure upgrade, and an important reference for investment location decisions. Previous incident hotspot identification methods are all region based, ignoring the underlying road network constraints. We first demonstrate how region based hotspot detection may be inaccurate. We then present Dijkstra’s-DBSCAN, a new network based density clustering algorithm specifically for traffic incidents which combines a modified Dijkstra’s shortest path algorithm with DBSCAN (density based spatial clustering of applications with noise). The modified Dijkstra’s algorithm, instead of returning the shortest path from a source to a target as the original algorithm does, returns a set of nodes (incidents) that are within a requested distance when traveling from the source. By retrieving the directly reachable neighbors using this modified Dijkstra’s algorithm, DBSCAN gains its awareness of network connections and measures distance more practically. It avoids clustering incidents that are close but not connected. The new approach extracts hazardous lanes instead of regions, and so is a much more precise approach for incident management purposes; it reduces the [Formula: see text] computational cost to [Formula: see text], and can process the entire U.S. network in seconds; it has routing flexibility and can extract clusters of any shape and connections; it is parallellable and can utilize distributed computing resources. Our experiments verified the new methodology’s capability of supporting safety management on a complicated surface street configuration. It also works for customized lane configuration, such as freeways, freeway junctions, interchanges, roundabouts, and other complex combinations.