Vehicle and engine emission tests are used to develop and certify compliance with regulation limits. Such tests produce a huge amount of complex experimental data. The use of Artificial intelligence can simplify and improve data analysis. This paper uses a random forest machine learning model on experimental data of ESC (European Stationary Cycle) tests done in a medium-duty diesel engine to discuss the importance of selecting the right input parameters to train the model and the model’s ability to predict engine operation regimes different of the ones used for training. The results of the study showed that the machine learning model of digital twins could accurately predict instantaneous fuel consumption in both steady-state and transient RDE emissions tests. The model was also able to predict tests other than those used for training, if the training data set used for training “covers” the one to be predicted.