An assessment of machine learning and data balancing techniques for evaluating downgrade truck crash severity prediction in Wyoming

Ampadu, Vincent Michael; Haq, Muhammad Tahmidul; Ksaibati, Khaled · 2022 · Crossref

DOI: 10.14254/jsdtl.2022.7-2.1

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 challenge of predicting crash severity for large trucks on downgrade roads in Wyoming, a region with high fatality rates due to steep grades, curves, and heavy truck traffic from oil and coal industries. Traditional regression models often fail to capture complex variable interactions, prompting the authors to evaluate the performance of seven machine learning (ML) models. The research aims to fill a gap in existing literature by comparing classification tree-based methods against non-tree-based algorithms to determine the most effective approach for severity prediction. The researchers utilized crash data from the Wyoming Department of Transportation spanning 2010 to 2019. The dataset included 120 attributes covering driver, vehicle, roadway, and environmental factors. Crash severity was categorized into five levels: Property Damage Only, Suspected Minor Injury, Possible Injury, Suspected Serious Injury, and Fatal Injury. To address potential multicollinearity, predictors were selected using a backwards stepwise procedure. The study compared four tree-based models—Adaptive Boosting (AdaBoost), Random Forest (RF), Gradient Boost Decision Tree (GBDT), and Extreme Gradient Boosting (XGBoost)—against three non-tree-based models: Support Vector Machines (SVM), Multi-layer Perceptron (MLP), and k-Nearest Neighbors (k-NN). The analysis also considered data balancing techniques to handle class imbalances inherent in crash severity data. The results demonstrated significant variation in model performance, measured by the Receiver Operating Characteristic Area Under the Curve (ROC AUC) score. The optimized Random Forest model achieved the highest accuracy with a ROC AUC of 95.296%. The k-NN model followed with 92.780%, while the MLP scored 87.817%. XGBoost performed moderately at 86.542%, whereas GBDT, SVM, and AdaBoost yielded lower scores of 74.824%, 72.648%, and 67.232%, respectively. Feature importance analysis identified the top ten predictors of crash severity, which included the use of safety equipment, airbag deployment, driver gender, and alcohol involvement. The study concludes that Random Forest is the superior method for predicting downgrade truck crash severity in Wyoming, outperforming other popular ML techniques. The findings highlight the importance of specific driver and vehicle factors, such as safety equipment usage and alcohol involvement, in determining crash outcomes. By demonstrating the efficacy of tree-based ensemble methods over traditional regression and other ML classifiers, the research provides a robust framework for transportation agencies to identify high-risk scenarios and develop targeted countermeasures to reduce severe truck crashes in mountainous regions.

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