Truck Accident Prediction and Risk Factors Analysis in Jordan: A Machine Learning Approach

Al-Tarawneh, Mu’ath · 2025 · Crossref

DOI: 10.25130/tjes.32.4.25

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 rising frequency and severity of truck accidents in Jordan, a problem exacerbated by rapid urbanization and a surge in vehicle ownership. While general traffic accident prediction has been studied globally, there is a lack of specific research focusing on truck drivers, who are involved in disproportionately fatal accidents due to unique operational factors. The research aims to identify key risk factors influencing accident involvement, frequency, and severity among truck drivers and to develop machine learning models capable of predicting these outcomes. To achieve this, the author developed a Driver Behavior Questionnaire (DBQ) comprising 24 questions covering demographic data, job characteristics, driver behaviors (e.g., distractions, speeding), and accident history. The survey was distributed to 988 male truck drivers operating along Jordan’s Desert Highway between October 2019 and January 2020. The data revealed that most drivers were aged 30–50, had 5–10 years of experience, and frequently engaged in distractions such as smoking (60% always) and using phones (12% always). Excessive speeding and poor road design were identified as the most common causes of accidents. The dataset was processed using WEKA software, with feature selection performed via the InfoGainAttributeEval method to identify significant predictors. To address class imbalance in the data, the Synthetic Minority Over-sampling Technique (SMOTE) was applied to generate balanced datasets for model training. The study constructed classification models to predict three targets: accident involvement (yes/no), number of accidents (0, 1–3, ≥4), and fatality occurrence (fatal/non-fatal). Feature selection indicated that driver behaviors (eating, smoking, radio use) and fatigue-related factors (sleep hours, travel distance, age) were significant predictors. The results demonstrated that using SMOTE-balanced data significantly improved model performance compared to imbalanced data. Specifically, the accuracy for predicting accident involvement increased from 75.4% to 84.5%; the accuracy for predicting the number of accidents rose from 69.7% to 85.5%; and the accuracy for predicting fatalities improved from 78.5% to 85.6%. The Decision Table classifier was identified as the most accurate model for accident involvement. The findings confirm that driver behavior and fatigue are critical determinants of truck accident risk and severity in Jordan. The study highlights the effectiveness of machine learning techniques, particularly when combined with data balancing methods like SMOTE, for accurate accident prediction. These results provide actionable insights for policymakers and the freight sector, suggesting that interventions targeting driver distractions, fatigue management, and speeding could mitigate accident rates. The proposed models offer a tool for assessing individual driver risk profiles, potentially guiding preventive measures and safety regulations in the Jordanian transport industry.

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