Driving behavior strongly influences both road safety and environmental sustainability. Aggressive patterns, such as rapid acceleration, harsh braking, and prolonged idling, can increase CO2 emissions by up to 40% and substantially increase accident risk. In contrast, eco-driving promotes smoother maneuvers that enhance both fuel efficiency and safety. Modern vehicles increasingly rely on telematics devices that record speed, distance, and location, allowing large-scale monitoring of driving styles. Emission modeling systems integrate these data with contextual variables (road grade, vehicle characteristics, weather) to improve prediction accuracy. The MOVES framework of the US Environmental Protection Agency is the current standard for emission estimation, but its direct use is limited. Recent work has introduced surrogate models, such as neuralMOVES, a deep neural network designed to replicate MOVES outputs with high fidelity. This study, conducted within the ECOSCORING project, a partnership between a major insurance company and the Federico II University, applies a pre-trained version of neuralMOVES to estimate CO2 emissions for 100,000 insured drivers over a two-month period. The analysis combines telematics data from onboard black boxes with customer and vehicle information supplied by the insurer. Emissions assessment alone does not foster behavioral change. To translate predictions into actionable guidance, we train ensemble models to explain emission estimates as a function of driver and vehicle characteristics. By integrating global and local model-agnostic interpretability techniques, we identify the specific driving behaviors that contribute the most to emissions and provide recommendations for safer, more sustainable practices. This framework illustrates how interpretability can bridge accurate deep learning models and practical decision support for insurers, thereby incentivizing eco-friendly driving while advancing broader environmental and safety goals.
6 Marzo 2026, ore 12
Alfonso Iodice D’Enza
Department of Political Sciences, University of Naples Federico II
In person: Room 34 (4th floor) building CU002 Scienze Statistiche
Webinar: https://uniroma1.zoom.us/j/83625004899?pwd=bXCtz0mp759PUh2lkqT0BUoVa0Uegg.1
ID riunione: 836 2500 4899
Passcode: 123456