Authors: Priyanka N Godiyal, Rutuja Amrale, Revati Ma’am, Archana Ma’am.
Abstract: Road traffic accidents are a leading cause of mortality worldwide, with India recording over 1.5 lakh fatalities annually. Identifying 'black spots' — specific road segments with disproportionately high accident frequency — is critical for targeted infrastructure intervention. Traditional methods of black spot identification rely on statistical thresholds applied to historical data, which are often reactive and location-agnostic. This paper proposes an integrated framework combining Machine Learning (ML) and Geographic Information Systems (GIS) for predictive black spot detection. We review and compare ML algorithms including Random Forest, XGBoost, Support Vector Machines (SVM), and Deep Neural Networks applied to multi-source data comprising accident records, road geometry, traffic volume, and environmental factors. Spatial analysis techniques such as Kernel Density Estimation (KDE) and spatial autocorrelation are used for feature engineering. Results show that ensemble methods achieve accuracy above 90%, with XGBoost yielding the highest AUC-ROC of 0.94. GIS-integrated output maps provide actionable, zone- specific risk rankings to support road safety planning.
DOI: https://doi.org/10.5281/zenodo.20118287
