Predicting vehicular emissions in downsized ethanol fueled flex-fuel vehicles using machine learning and real-world driving data

Laboratory emission certification has historically been the standard approach for vehicle homologation. However, since 2022, Brazilian legislation has incorporated Real Driving Emission (RDE) testing, with further restrictions introduced under the PROCONVE L8 phase. These regulatory developments have increased the demand for scalable, cost-effective emission monitoring tools. At the same time, ethanol fueled powertrains have emerged as a strategic pathway for greenhouse gas mitigation in countries where full fleet electrification remains economically challenging. This study proposes a machine learning-based emission prediction framework using OBD-II (On-Board Diagnostic II) and tailpipe gas flow vehicle operational data acquired during RDE testing in a flex-fuel vehicle with Portable Emission Measurement Systems (PEMS). The objective is to evaluate whether simplified and non-invasive data acquisition can provide reliable estimates of CO, NOx, and total hydrocarbon (HCt) emissions in ethanol-fueled flex-fuel vehicles. Four machine learning algorithms were evaluated, with eXtreme Gradient Boosting (XGBoost) achieving the best overall performance. The model obtained R² values of 0.9171, 0.8559, and 0.8661 for CO, NOx, and HCt, respectively. These results demonstrate that emission prediction based on OBD-II data and tailpipe flow is feasible and can support fleet management strategies, and regulatory screening applications. The findings highlight the potential of low-cost machine learning solutions to complement traditional emission measurement methodologies, particularly in emerging mobility markets with strong biofuel penetration.