Applying Data Mining Methods to Explore Animal-Vehicle Crashes

Abstract

Animal-vehicle crashes (AVCs) are a significant issue in Louisiana that requires attention. Data on AVCs that occurred from 2015 to 2020 were obtained from the Louisiana Department of Transportation and Development (DOTD), including 14,349 crashes with major injury (KA), minor injury (BC), and no injury (O) severity groups. Aiming to find the collective association of attributes from AVC data, which are categorical in nature, this study utilized two data mining methods: multiple correspondence analysis (MCA) and association rule mining (ARM). Five hierarchical clusters that were generated from the BC and O AVC datasets were particularly significant. Among several other findings, MCA revealed that BC and O AVCs are more concentrated on parish roads during the spring season, while O AVCs in the fall and winter tend to occur on highways with speed limits of 50 mph or higher. ARM revealed that moderate-speed parish roads are frequently associated with KA and BC AVCs, particularly in residential areas and during the spring season, and they often involve young drivers. The findings of this study can be particularly beneficial by considering the spatiotemporal factors associated with animal concentration and movement to develop targeted interventions and mitigation strategies.