A persistent challenge in using naturalistic driving data is identifying events of interest from a large data set in a cost-effective manner. A common approach to this problem is to develop kinematic thresholds against which kinematic data is compared to identify potential events or kinematic triggers. Trained video analysts are then used to determine if any of the kinematic triggers have successfully identified events of interest. Video validation for a large number of kinematic triggers is time consuming, expensive, and possibly prone to error. The use of video analysis to review a large number of kinematic triggers is due to an inability to effectively discriminate between innocuous driving situations and safety-relevant events in an automated manner. A potential solution to this problem is the development of classification models that would reduce the number of kinematic triggers submitted for video validation through a process of pre-validation trigger classification. A functional yaw rate classifier was developed that retains a majority of safety relevant events (92% of crashes, 81% of near-crashes) while reducing the number of invalid or erroneous yaw rate triggers by 42%. For large-scale studies such a reduction in the number of invalid triggers submitted for video validation allows video analysis resources to be focused on the aspect of driving research in which it add the greatest value: providing contextual information that cannot be derived from kinematic and parametric data.