Using Random Forest to Test If Two-Wheeler Experience Affects Driver Behaviour When Interacting with Two-Wheelers

Car drivers are primarily responsible for crashes between cars and bicycle and motorcycle riders (two-wheelers; TWs). A lack of exposure and riding experience with TWs among car drivers may contribute to the occurrence of these crashes. The current research investigates if car drivers with different TW riding experience levels act differently during risky interactions with both types of TWs. A total of 69 drivers completed a 10-minute driving session using the CARRS-Q advanced driving simulator, which included five interactions with TWs based on common crash types. These interactions involve driving manoeuvres in which TWs were initially positioned in front of or at a right angle to the driver. The drivers were divided into two categories based on whether they possessed TW riding experience or not. For analysis, several response features were calculated using the speed and time to collision (TTC) during the time window covering the interaction. Then, a random forest algorithm used the extracted features from the predefined window to classify the drivers. Classification accuracy was calculated using leave one out cross-validation. To test the association between the speed, TTC features and experience as a TW rider, the drivers’ labels were randomly permuted. The random forest models were trained using these permuted labels, and the average classification accuracy was calculated. Overall, the comparison between the average of the original data classification accuracy and the randomly permuted data revealed a statically significant difference. This study found that two-wheeler experience influences driver behaviour when interacting with two-wheelers. Further research should investigate whether additional driver education, potentially in the form of online training during the licensing process, would increase awareness of two-wheelers and improve the safety of interactions with such road users.