Fuel consumption in companies in the rail transport sector represents one of the largest operating expenses and one of the biggest concerns in terms of pollutant emissions. The high fuel consumption also entails a high representation in the emissions scope matrix (more than 90 percent of railroad emissions come from fossil fuel consumption). Aiming to seek constant operational improvement, numerous studies have been carried out proposing new tools to reduce fuel consumption in the operation of a freight train. In this way, it is important to highlight the improvement of train driving parameters that can be calibrated to reduce fuel consumption. To accomplish this goal, the present work implements two machine learning models to predict the energy efficiency of a freight train: random forest and artificial neural networks. The random forest achieves the best performance against the models, with an accuracy of 91 percent. To calculate how much each parameter influences the prediction model, this work also uses the technique of accumulated local effects for each parameter related to energy efficiency. The final results show that, within the four analyzed calibration parameters, the traction per transported ton indicator presented greater representation in terms of absolute impact on the energy efficiency of a freight train.