A hybrid data mining framework to investigate roadway departure crashes on rural two-lane Highways: Applying Fast and Frugal Tree with Association Rules Mining

Abstract

The complexity of factors contributing to roadway departure (RwD) crashes on rural highways necessitate advanced analytical approaches to enhance traffic safety. This study presents a hybrid data mining framework that combines the Fast and Frugal Tree (FFT) and Association Rules Mining (ARM) algorithms to identify the patterns of RwD crashes on rural 2-lane highways in Louisiana state. The research is focused on addressing two critical research questions (RQ), RQ1: Which variable features contribute to the fatal-severe RwD crashes? RQ2: Focusing on the identified top factors contributing to fatal-severe RwD crashes, how co-occurrence of different crash-contributing factors increase the likelihood of RwD crashes? For the analysis, this research team collected crash data from the Louisiana Department of Transportation and Development, encompassing a total of 22,406 unique RwD crashes on rural 2-lane highways. In the first stage (addressing RQ1), the FFT model identified the top variable features contributing to fatal-severe RwD crashes, including no use of seatbelts, alcohol-impaired driver condition, male gender, dark-no-streetlight, older drivers (>64 years), 12 am – 6 am, light truck, and so on. In the second stage (addressing RQ2), based on the factors identified by FFT, ARM explored how these factors interact and associate, revealing intricate drivers’ behavioral patterns related to RwD crashes. This comprehensive analysis uncovers not only the individual impact of these factors but also their combined effects, offering a deeper understanding of the dynamics of RwD crashes. This research contributes valuable insights into evidence-based, data-driven strategies to reduce the frequency and severity of RwD crashes on rural highways, advancing traffic safety initiatives and improving safety on rural 2-lane roadways.