Abstract :

Traffic Accident Severity Prediction Supports Emergency Response And Road Safety Policy, But Real-world Adoption Requires Reliable Performance And Transparent Explanations. This Paper Presents An End-toend, Reproducible Machine Learning Pipeline For Multi-class Accident Severity Prediction Using A Large-scale Crash Dataset. The Workflow Includes Schema Validation, Missing-value Handling, Stable Categorical Encoding, Temporal Feature Extraction, Class-imbalance-aware Learning, Cross-validated Model Selection, And Explainability Using SHAP. We Benchmark Three Gradient-boosted Tree Ensembles (XGBoost, LightGBM, CatBoost) Using Balanced Accuracy, Macro Precision/recall, And Macro F1-score. LightGBM Achieves The Best Overall Results (balanced Accuracy 0.7028, Macro F1 0.4054). System Testing Is Performed Across Pipeline Integrity, Preprocessing Stability, Performance Reliability, And Explainability Consistency To Validate Correctness And Reproducibility. The Resulting Framework Provides Both Strong Predictive Baselines And Actionable Interpretations For Severity-specific Safety Interventions. Keywords— Traffic Safety Analytics; Accident Severity Prediction; Gradient Boosting; LightGBM; Class Imbalance; Explainable AI; SHAP; System Testing.

Published:

03-2-2026

Issue:

Vol. 26 No. 2 (2026)

Page Nos:

12-18

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite