Effect of Sampling Frequency and Window Size on Identifying Risky Driving Maneuvers

The identification of risky driving maneuvers, such as harsh acceleration, harsh braking, harsh cornering, and speeding, has always been an important topic in monitoring driving safety, which lays the groundwork of a wide range of real-world applications ranging from personal car insurance to risk management in fleets. Although extensive research has been carried out on the identification of risky driving maneuvers using various driving trajectory datasets, to the best of our knowledge, there is not yet a thorough study on the effect of the sampling frequency of the telematics devices and the window size used to efficiently identify risky driving maneuvers. In this paper, we focus on the evaluation of the effect of sampling frequency and window size on the identification of risky driving maneuvers. Specifically, we first implement an enhanced multi-stage approach named SCALE to identify risky driving maneuvers, comprising the following procedures: Sampling raw data, Clustering sampled data, lAbeLing clustered data, and training a classifiEr using the labeled data. We then apply SCALE to extensively evaluate the effect of different sampling frequency rates and window sizes using a real-world comprehensive driving trajectory dataset. Experimental results suggest that a sampling frequency of 5 Hz with a window size of 3 s would be sufficient to capture a vast majority of risky driving maneuvers.