IMPROVING ROAD SAFETY: SUPERVISED MACHINE LEARNING ANALYSIS OF FACTORS INFLUENCING CRASH SEVERITY

Khanh Giang, LE · 2025 · DOAJ

DOI: 10.20858/sjsutst.2025.127.8

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 critical need for interpretable, region-specific analysis of road traffic crash severity, focusing on the United Kingdom. While machine learning (ML) has been widely applied to road safety, existing research often lacks transparency in model decision-making and specific insights for the UK context. The authors aim to bridge this gap by evaluating the predictive performance of three supervised ML models—Decision Tree (DT), Support Vector Machine (SVM), and LightGBM—and employing SHapley Additive exPlanations (SHAP) to identify the key factors influencing crash outcomes. The research utilizes a dataset of 106,004 road traffic accident records from the UK Department for Transport for the year 2022. The data includes variables related to road conditions, environmental factors, vehicle involvement, and police attendance. The methodology involves rigorous data preprocessing, including the removal of redundant features and encoding of categorical variables. Feature selection was performed using a Random Forest algorithm, retaining attributes that accounted for 90% of cumulative importance. The dataset was split into 80% training and 20% testing sets, with 10-fold cross-validation employed to ensure robustness. The models were trained to classify accident severity into three categories: fatal, serious, and slight. Performance was evaluated using accuracy, precision, recall, F1-score, and confusion matrices. The results demonstrate that the LightGBM model outperformed both the Decision Tree and Support Vector Machine in predictive accuracy. Through SHAP analysis, the study identified police officer attendance at the scene, speed limits, and the number of vehicles involved as the most pivotal determinants of crash severity. Specifically, higher speed limits and single-vehicle collisions were found to correlate with more severe outcomes. Conversely, the presence of a police officer at the scene appeared to mitigate accident severity. The analysis provides a clear ranking of feature importance, offering actionable insights into which variables most significantly impact the classification of crash severity. The significance of this work lies in its integration of high-performance ML with post-hoc interpretability techniques, addressing the "black box" limitation often associated with AI in safety-critical domains. By identifying specific, actionable factors such as speed regulations and enforcement strategies, the findings offer valuable guidance for policymakers aiming to reduce traffic-related fatalities and injuries. The study underscores the potential of interpretable ML frameworks to enhance the understanding of crash dynamics and inform targeted safety interventions, contributing to broader global efforts in road safety improvement.

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