Hybrid vehicles, which combine an internal combustion engine with an electric motor, serve as a viable alternative in regions where infrastructure and electricity production contribute significantly to CO2 emissions. The paper presents an approach that makes it possible to reduce the efforts to analyze and the number of expensive vehicle emission tests. A digital twin, trained by machine learning algorithms, was applied to a FTP75 emission test data of a Hybrid vehicle when both internal combustion engine and electrical motor are used to move the vehicle. The developed Random Forest model was able to predict with high accuracy the instantaneous CO2, THC and CH4 emissions for a Hybrid vehicle on the FTP75 cold phase. Deviations between the model and the measured accumulated values at end of test, were lower than 1%. The effect of exhaust catalyst temperature and other powertrain parameters was discussed. The model’s accuracy depends on the correct selection of the input parameters and that the “experimental space” of the simulated cycle is relatively well covered during training.