| Authors | DR. JYOTI. Y. KULKARNI |
|---|---|
| Article Type | Research Article |
| Language | English |
| Journal | North Asian International Research Journal of Multidisciplinary |
| ISSN | 2454-2326 |
| Volume | 12 |
| Issue | 5 |
| Pages | 29-41 |
| Publication Year | 2026 |
| Publication Date | May 01, 2026 |
| DOI URL | https://doiglobal.org/10.2026/NAIRJCMD.010 |
Road traffic crashes remain a major transportation and public-safety challenge in India. Predicting crash severity can support proactive engineering, enforcement and emergency-response decisions by identifying combinations of road, vehicle, environmental and temporal factors associated with serious outcomes. This paper proposes an explainable machine-learning framework for predicting accident severity on Indian highways. The framework integrates structured crash records, data cleaning, categorical encoding, imbalance treatment, feature engineering, Random Forest, Gradient Boosting and XGBoost-style tree ensembles, followed by SHAP-based interpretation. Rather than presenting fabricated experimental performance, the paper uses published Indian highway studies as comparative evidence and defines a reproducible protocol for future model validation. Apeer-reviewed Indian-highway Random Forest study using 3,257 observations and 32 variables reported 67% training accuracy but only 41.47% test accuracy, demonstrating the danger of imbalance and weak generalization. A 2025 MethodsX framework subsequently emphasized Random Forest, Gradient Boosting and SHAP interpretation for Indian highway severity prediction. Another Indian expressway study analyzed 2,747 crashes using multinomial logit, decision tree and Random Forest models. The proposed research therefore prioritizes balanced evaluation, macro-F1, class-wise recall, calibration and explainability rather than accuracy alone. The paper argues that interpretable AI can help identify high-risk conditions and support road-safety planning, but predictive models should supplement not replace engineering judgment, police investigation and validated crash databases.
Total Views
Total Downloads
Publication Year
DOI Status
This scholarly article has been successfully registered and assigned a persistent DOI through DOIGLOBAL.
DOI: 10.2026/NAIRJCMD.010
Journal:
North Asian International Research Journal of Multidisciplinary
DOI Provider:
DOIGLOBAL
This scholarly record is permanently registered through DOIGLOBAL DOI Infrastructure and remains accessible using its persistent DOI identifier.