An interpretable, data-driven framework empowered by explainable AI for fuel consumption and CO2 emission prediction

초록

The growing concern over environmental sustainability and regulatory demands necessitates a deeper understanding of fuel consumption and CO2 emissions. However, accurate prediction remains challenging because the impact of several factors can be intertwined, and the results of predictive models are not easily interpretable. Conventionally used approaches often rely on black-box or oversimplified linear models, thereby failing to capture the complex relationships embedded in high-dimensional data. These result in a noticeable lack of actionable analytical insights that may be used to guide decision-making and policymaking; hence, their use is limited to policymakers, environmental scientists, and industries. This limited interpretability significantly restricts the adoption of such models in real-world sustainability planning and emission control strategies. Explainable AI (XAI) offers a robust and transparent solution to bridge this critical gap. Unlike conventional machine learning models, which often lack transparency for end-users, XAI combines high levels of accuracy with explainability. As a result, this research proposes an enhanced interpretable, data-driven framework empowered by XAI to overcome these limitations. The framework was evaluated using a publicly available Kaggle dataset comprising 639 vehicle samples and achieved strong predictive performance with R2 value of 0.9168 and RMSE of 14.93 using the Extra Trees Regressor model. To ensure transparency and insight-based interpretation, the proposed model adopts contemporary XAI approaches to explain fuel consumption and CO2 emissions at the instance level. This method supports informed decision-making by identifying key emission determinants, quantifying their relative influence, and revealing the underlying relationships between vehicular attributes and emission outcomes. Compared to previous approaches, the proposed framework demonstrates superior predictive accuracy, improved interpretability, and enhanced practical applicability, establishing it as a reliable solution for sustainable transportation management and data-driven environmental policymaking.

키워드