Fuel Efficiency Modeling and Prediction for Automotive Vehicles: A Data-Driven Approach

This study is mainly concerned with fuel efficiency modeling and prediction for common automobiles based on an informative vehicle database. The historical database is processed and the mutual information index (MII) is employed to identify a set of characteristics that significantly affect fuel efficiency. Five different machine learning techniques are exploited to build fuel efficiency prediction models. Among these techniques, quantile regression, which is a natural extension of classical least square estimation, is shown to have better performance for fuel efficiency prediction compared to other adopted techniques. It is also demonstrated that with the selected attributes based on MII, the prediction performance is almost ideal when exploiting the complete dataset.