Investigating patterns of freeway crashes in Jordan: Findings from a text mining approach

Abstract

Effective road safety measures rely on understanding the trends and factors influencing traffic accidents. This study employs a text-mining approach to analyze crash narratives from 7,587 crash records on five major Jordanian freeways between 2018 and 2022. By applying methods such as Word Co-occurrence Network (WCN), Rapid Automatic Keyword Extraction (RAKE), Probabilistic Topic Modeling (LDA), and Association Rule Mining (ARM), the analysis uncovered key insights into traffic crash dynamics. Key findings reveal that violations (with lift values exceeding 1.5 in ARM) and mechanical failures, such as vehicle malfunctions and tire blowouts, were major contributors. Environmental factors, including oil leaks and stray animals, were also significant triggers. Additionally, high-risk behaviors like sudden lane changes and non-compliance with traffic rules were identified. Recommendations include infrastructure improvements, driver education, and targeted measures to mitigate animal-related crashes. This study is among the first in developing countries to use advanced text mining techniques for freeway crash narratives, addressing a critical research gap.