Estimation of Road Accident Severity Using K-Means Clustering of Spatio-Temporal Data with Backend Spatial Data Infrastructure

Road accidents claim the lives of over four hundred individuals daily in India. While several researchers have focused on identifying accident-prone locations using different parameters and machine learning models, current solutions often lack a comprehensive framework for identifying blackspots based on recent fatality information. In this paper, we present a geospatial framework for identifying accident blackspots, considering multiple complex parameters such as accident frequency, total fatalities, fatal accidents, fatal and major accidents, and Equivalent Property Damage Only (EPDO). We employ an upper-tail critical test on each parameter to calculate the blackspots. Additionally, we group the blackspots using k-means clustering based on the severity of the accident locations within the spatial data infrastructure. We consider four clusters representing different zones, and each cluster is differentiated by its risk factor, determined by accident density. The blackspot identification is conducted on all roads in Kolkata, West Bengal, India. The zones are ranked in increasing order of risk factors as west, east, north, and south. The proposed model is replicable, scalable, and applicable to any road network.