SPEED-RELATED TRAFFIC ACCIDENT ANALYSIS USING GIS-BASED DBSCAN AND NNH CLUSTERING

Abstract

Abstract. To ensure road safety and reduce traffic accidents, it is essential to determine geographical locations where traffic accidents occur the most. Spatial clustering methods of hot spots are used very effectively to detect such risky areas and take precautions to minimize or even avoid fatal or injury accidents. This study aims to determine speed-related hot spots in the pilot Mersin Region, which includes seven cities in the central-southern part of Turkey. Two different hot spot clustering methods, the Nearest Neighbourhood Hierarchical Clustering Method (NNH) and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) Method, were employed to analyse traffic accident data between 2014-2021, obtained from the General Directorate of Highways. CrimeStat III program, which is free software, was used to detect NNH clusters, while the DBSCAN clusters were obtained using the open-source GIS program QGIS, which was also used to visualize and evaluate the results comparatively. As a result of the analysis, it was determined which method gave more effective results in determining the traffic accident risk clusters. These clusters were analysed based on road geometries (intersection or corridors). In addition, by considering the areas where repeated accidents have occurred over the years, future predictions of traffic accidents have been estimated.