Application of machine learning to severe injury prediction in rural run-off-road crashes
This paper describes how Machine Learning (ML) techniques were used gain a deeper insight into the factors leading to rural run-off-road casualty crashes being severe. The ML findings were compared with a conventional binary logistic modelling approach. The findings showed that roadside objects hit, road curvature, vehicle type and age, and the number of persons in vehicle were strong predictors of run-off-road crash severity. More importantly, ML highlighted specific combinations of risk factors which were linked to high risk of severe injury in a run-off-road casualty crash. ML may enable a more synergistic approach to risk and Safe System assessment.