An Intelligent Model for Traffic Accident Hotspot Detection

Abstract

Identifying traffic accident hotspots is essential for improving urban road safety and supporting Data-driven transportation planning. Despite the availability of extensive traffic data, many existing hotspot detection approaches are constrained by rigid spatial assumptions and limited integration of contextual risk factors. This study proposes a comprehensive framework for accident hotspot detection and risk characterization in Los Angeles using the 2023 U.S.  accidents dataset. After data cleaning and feature engineering, the data were analyzed using the Density Based Spatial Clustering with Noise (DBSCAN) algorithm. The model identified 12 distinct hotspots, for each hotspot, a composite Danger Score was computed by combining normalized measures of accident frequency, severity, nighttime incidence, and adverse weather conditions. These scores were then partitioned into three risk levels High (n = 4), Medium (n = 4), and Low (n = 4) using k means clustering. Cluster validity was assessed using the Silhouette Score (0.565), Davies–Bouldin Index (0.439), and Calinski–Harabasz Index (2121.3), confirming a clear and robust spatial structure. The results show that reduced visibility, nighttime driving, and localized weather variability particularly in high density freeway and arterial segments substantially contribute to increased accident risk. The proposed framework enables transportation agencies to proactively identify hazardous locations and prioritize targeted safety interventions in Los Angeles and other large metropolitan regions.