TinyML Implementation and Optimization for Fuel Type Classification on OBD-II Edge Device

Global concern over climate change has spurred a demand for precise methods to quantify Greenhouse Gas (GHG) emissions in the transportation sector, where emissions from fossil fuel use are significant. Conventional methods for determining vehicle fuel types often rely on manual and declarative processes, introducing inaccuracies that hinder efficient environmental policy implementation and the creation of reliable carbon inventories. Therefore, this article proposes a machine learning-based approach for the automated identification of vehicle fuel types, utilizing data collected from vehicle sensors and the Random Forest technique implemented in TinyML. This approach, with its practicality and potential to be implemented in a wide range of vehicles, offers a promising solution to the industry's need for accurate carbon inventories. Additionally, a case study was conducted using the OBD-II Edge Freematics ONE+ device, enabling real-time data collection and machine learning model execution directly on the device. The study evaluated the model's accuracy in a real-world scenario. The approach demonstrated high precision rates in identifying fuel types, underscoring the viability of TinyML for classification applications in resource-constrained embedded devices. The findings suggest that this approach can contribute to environmental management in the transportation sector by facilitating the development of precise carbon inventories and supporting effective emission reduction policies.