One of the main environmental concerns lies in greenhouse gas emissions originating from vehicles, particularly carbon dioxide (CO2). The transition to cleaner vehicles and the consequent reduction of these emissions are emerging as global trends. In this context, the Internet of Things (IoT) and the OBD-II (On-Board Diagnostics) protocol play a crucial role in real-time monitoring and analysis of vehicle emissions. This study proposes a methodology that utilizes online unsupervised learning algorithms, applicable to the context of TinyML (Machine Learning on low-power devices), to estimate CO2 emissions and perform a comparative analysis between a vehicle fueled by ethanol and gasoline. The results showed that CO2 emissions from ethanol were significantly lower compared to gasoline. The study also analyzed the evolution of these emissions over time and at different speed ranges, as well as their variations along the route.