Abstract:
|
Latest advancements in the Internet of Things domain enable live transmission of data that can be used to extract information on driver behavior and facilitate the adoption of new insurance models. Although ordinary and commercial drivers are subject to different driving settings and requirements, previous studies have mainly focused on developing insurance models for ordinary drivers. The present paper analyzes naturalistic driving data from the commercial operations of 3,854 drivers, generated in 2018 from one of the first vehicular applications of Long Range technology. After comparing five classification algorithms, Gradient Boosted Trees resulted in the highest area under the precision and recall curve (0.438). Further, running driving time (non-idling), rapid speed changes and trip frequency resulted as the most influential variables on the probability of collision. Business-related variables (running driving time and trip frequency) can inform fleet companies on how to instate rental rates for customer businesses, whereas variables indicating risky driver behaviors (rapid speed changes) can inform the design of real-time feedback systems aimed at correcting such behaviors.
|