To train the model, we selected a mean squared error loss function that penalizes large deviations.