Using unsupervised learning to investigate injury-associated factors of animal-vehicle crashes

Animal vehicle crash is a critical yet often under-emphasized safety concern of Louisiana. During 2014-2018, over 14,000 animal-related crashes cost Louisiana more than $520 million. To identify multiple key contributing factors and their association patterns, this study applied association rules mining in the dataset of animal-related roadway crashes that occurred during 2014-2018. Since high proportions of animal-related crashes involve complaint and no injury of vehicle occupants, separate analyses were performed for KAB (fatal, severe, and moderate injury) and CO (possible/complaint and no injury) crashes. Top rules ordered by higher lift values were interpreted and compared to implicate the quantified likelihood of crash patterns. KAB rules presented the likelihood of associations of characteristics such as unlighted dark conditions, interstate and parish roads, a wide range of speed limits, residential and open country locations, normal and rainy weather conditions, light trucks, young drivers, etc. The majority of CO crash patterns were associated with interstates, straight segments, normal driver conditions, clear weather, unlighted dark conditions, open country locations, a speed limit of 97 km/h or higher, etc. Findings in this study and their implications supported by prior studies are expected to be beneficial in strategic planning for identifying implementable countermeasures for animal-vehicle crashes.