Using Machine Learning Models to Forecast Severity Level of Traffic Crashes by R Studio and ArcGIS

Al-Mistarehi, Bara’ W.; Alomari, Ahmad H.; Imam, Rana; Mashaqba, Mohammad · 2022 · OpenAlex-citations

DOI: 10.3389/fbuil.2022.860805

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 improved traffic safety measures by forecasting the severity levels of traffic crashes using machine learning (ML) and Geographic Information Systems (GIS). Motivated by the rising number of accidents and significant economic and social burdens in Jordan, the research aims to identify crash causes, risk factors, and accident hot spots to develop effective countermeasures. The study focuses on Zarqa City, a major urban area in Jordan with high accident frequency, utilizing a dataset of nearly 97,900 traffic accidents recorded between 2014 and 2018. The data includes diverse variables such as driver characteristics, vehicle types, road geometry, weather conditions, and injury severity levels (slight, medium, severe, and fatalities). The methodology combines supervised and unsupervised ML algorithms with spatial analysis. Supervised learning techniques, specifically Random Forest, Decision Tree, and AdaBoost, were employed to predict crash severity based on road crash elements. These models were trained and tested using a 70/30 data split, with performance evaluated via accuracy, precision, recall, F1 score, and error rates. Additionally, an unsupervised association rule algorithm was used to uncover relationships between driver behavior, highway geometry, environmental factors, and injury outcomes. Spatial statistics, including Optimized Hot Spot Analysis, were applied to map high-density accident locations and fatality clusters. The results demonstrate that the Random Forest model significantly outperformed the Decision Tree and AdaBoost algorithms across all injury categories. Random Forest achieved the highest accuracy rates, ranging from 97.93% for slight injuries to 99.96% for fatalities. It also recorded the highest precision and F1 scores, with precision reaching 100% for medium, severe, and fatal injuries. In contrast, the Decision Tree model showed lower accuracy (approximately 33%) and higher error rates, although it maintained high recall values. The association rule analysis identified specific variable combinations that significantly impact injury severity, while spatial analysis successfully mapped accident hot spots, revealing correlations between crash density and specific road or environmental features. The significance of this research lies in its validation of Random Forest as the most suitable algorithm for predicting traffic crash severity in urban environments. By accurately forecasting injury levels and identifying high-risk locations, the study provides transportation planners and policymakers with data-driven insights to implement targeted safety interventions. The integration of ML with GIS offers a robust framework for understanding complex behavioral and environmental factors contributing to accidents, ultimately supporting efforts to reduce fatalities and improve road safety standards in Jordan and similar 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