Binary and Multi-Class Crash Severity Prediction: A Cross-Algorithm Analysis

Document Type

Conference Proceeding

Publication Date

1-1-2026

Abstract

Traffic crash severity prediction faces significant challenges from class imbalance and limited model interpretability as critical concerns for the safety-critical applications that require a transparent and non-biassed decision making. This paper presents a comprehensive analysis of artificial intelligence techniques for binary and multi-class crash severity prediction using 745,326 traffic accidents collected from the FARS database covering all US states from 2015 to 2023. We systematically compared multi-machine learning algorithms: XGBoost, CatBoost, Easy Ensemble, and Random Forest on two-class (fatal and no fatal) and three-class severity classification (fatal, potential, and no injuries). Our investigation reveals all models achieved remarkably similar performance (83% accuracy and 0.82-0.83 F1-score) despite employing different mathematical approaches. Key contributions include documenting cross-algorithm performance convergence and implementing AI techniques for safety applications. These findings offer valuable insights for deploying reliable prediction systems in transportation safety management.

Publication Title

International Conference on Transportation and Development 2026 Transportation Safety and Emerging Technologies Selected Papers from the International Conference on Transportation and Development 2026

ISBN

[9780784487013]

Link to Full Text

Share

COinS