Crashes classification in naturalistic driving scenarios using random forest machine learning algorithm

This research analyses the viability of utilising observed kinematics in machine learning models to identify safety critical events (SCE’s). There is a need for efficient algorithms to idenfitify SCE’s in large datasets, such as naturalistic driving studies (NDS). Typical threshold approaches, while fast, often fail to distinguish between normal driving and SCE’s. The methodology proposed presents strong evidence that kinematic features used in machine learning models have good classification capabilities