Understanding the drivers of deer-vehicle collisions in South Carolina, U.S.A.

This is the authors' abstract. We don't add key points for this paper.

Every year in the U.S.A., approximately 1.5–2.1 million deer-vehicle collisions (DVCs) occur, resulting in 200–440 human fatalities, 26,000–59,000 human injuries, and greater than $10 billion in property damage. However, there is a lack of studies implementing machine learning (ML) techniques at the state level to evaluate factors affecting DVCs. Data on DVCs are valuable to reduce the occurrence of DVCs and to assist in planning. We utilized DVC location data from 2018 to 2021 provided by the U.S.A. Department of Transportation and Safety to identify DVC factors. Findings suggest that DVCs occurred more frequently near developed areas, cultivated land and woody wetlands, and during October from 8:00 PM to 10:00 PM and 6:00 AM to 8:00 AM. Result from negative binomial regression suggests that counties with higher hog harvest, coyote harvest, and higher amounts of water are more prone to DVCs. The hotspot analysis revealed that hot spots for DVCs shifted from 2015 to 2023. The ML models found accuracy scores of 0.67 and 0.66 from random forest and artificial neural networks, respectively. These results provide a foundation for future models that incorporate more detailed, dynamic predictors.