The actual driving condition and fuel consumption rate gaps between lab and real-world are becoming larger. In this paper, we demonstrate an approach to determine the most important factors that may influence the prediction of real-world fuel consumption rate of light-duty vehicles. A multilayer perceptron (MLP) method is developed for the prediction of fuel consumption since it provides accurate classification results despite the complicated properties of different types of inputs. The model considers the parameters of external environmental factors, the manipulation of vehicle companies, and the drivers' driving habits. Based on the BearOil database in China, 2,424,379 samples are used to optimize our model. We indicate that differences exist between real-world fuel consumption and standard fuel consumption under simulation conditions. This study enables the government and policy-makers to use big data and intelligent systems for energy policy assessment and better governance. INDEX TERMS Artificial intelligence, big data, multilayer perceptron, fuel consumption rate, light-duty vehicles.