DBSCAN-Based Spatial Hotspot Detection of Motorcycle and Three-Wheeled Vehicle Crashes

Abstract

Because of the lack of physical protection and the often-complex interaction between motorcycles and 3-wheelers and the other components of urban transportation systems, motorcycles and 3-wheelers are among the most vulnerable road users. Therefore, it is important to determine spatial crash hotspots to enable evidence-based road safety management and targeted interventions. The purpose of this study is to detect crash hotspot areas of motorcycles and three-wheel vehicles for the Rusafa district in Baghdad using the density-based spatial clustering of applications with noise (DBSCAN) algorithm and the Haversine distance measure. There were 802 crash records that were georeferenced and analyzed to identify areas that have been more prone to crashes. We used a Ball Tree nearest-neighbor search algorithm to implement DBSCAN and preprocessed the geographic coordinates before we applied DBSCAN, converting them to radians. We did a sensitivity analysis and selected the optimal clustering parameters (eps = 0.5 km, min_samples = 5). Three spatial indicators (Zone_Label, Zone_Count, Hotspot_Flag) were then calculated to describe the identified hotspot regions. The framework recommended 37 spatial clusters, with 656 (81.8%) of these crashes being identified as a hotspot and 146 crashes (18.2%) as spatial noise. The greatest number of crashes recorded occurred in the largest hotspot (Zone 8) with 116 crashes, or 14.46% of all crashes recorded. Results also indicate that the most common crash locations are not necessarily the most severe, suggesting that, in addition to severity, crash frequency should be taken into account when assessing road safety. The framework proposed here may assist transportation agencies in determining persistent crash locations, prioritizing engineering and enforcement measures, and combining with machine learning methods to forecast crash severity