Comparative Analysis of Parametric and Non-Parametric Data-Driven Models to Predict Road Crash Severity among Elderly Drivers Using Synthetic Resampling Techniques

Alrumaidhi, Mubarak; Farag, Mohamed M. G.; Rakha, Hesham A. · 2023 · Crossref

DOI: 10.3390/su15139878

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the growing public health concern of road crash severity among elderly drivers (aged 65 and older), a demographic increasingly vulnerable to severe injuries and fatalities due to age-related physical and cognitive declines. The research aims to predict crash severity using machine learning models while addressing the critical challenge of class imbalance in crash datasets, where severe crashes are significantly underrepresented compared to non-severe ones. The authors specifically investigate whether synthetic resampling techniques can enhance the predictive performance of both parametric and non-parametric models. The analysis utilized crash data from the Virginia Department of Transportation spanning 2014 to 2021, focusing on a subset of 157,800 crashes involving senior drivers. Crash severity was categorized into non-severe (injury levels O, B, C) and severe (K, A) using the KABCO scale. The study compared four machine learning models: two parametric models (logistic regression and linear discriminant analysis) and two non-parametric models (random forest and extreme gradient boosting). To mitigate class imbalance, the researchers applied two synthetic resampling techniques: Random Over-Sampling Examples (ROSE) and the Synthetic Minority Over-Sampling Technique (SMOTE). The dataset was partitioned into training (70%) and test (30%) sets using stratified sampling, and model performance was evaluated using metrics including accuracy, sensitivity, specificity, balanced accuracy, and geometric mean. The results demonstrated that synthetic resampling significantly improved model performance. For parametric models, the inclusion of resampling techniques increased the true positive rate for severe crash prediction in logistic regression from 6% to 60% and boosted the geometric mean from 25% to 69%. Similarly, SMOTE improved the true positive rate for the XGBoost model from 8% to 36%. Notably, the study found that parametric models outperformed non-parametric counterparts when balanced resampling techniques were employed. Beyond prediction, the analysis identified key contributing factors to crash severity, providing insights into the variables that heighten risk for elderly drivers. The significance of this work lies in its demonstration that synthetic resampling is essential for accurate crash severity prediction in imbalanced datasets, particularly for vulnerable populations. By establishing the superiority of parametric models under these conditions and identifying specific risk factors, the study offers actionable insights for developing targeted interventions. These findings support the development of effective safety strategies, policy decisions regarding elderly driver licensing, and infrastructure designs that accommodate the needs of older drivers, ultimately aiming to reduce the burden of severe crashes on healthcare systems and economies.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: crash risk outcomes