Results from three different experiments to estimate the performance of the trained models generalising on unseen data. Models trained on Mg II h&k perform best in all experiments, followed by the models trained on C II and then Si IV. Contrary to Si IV and C II the performance on each dataset remains similar in the first two experiments, while the models trained on Si IV and C II always perform better in the cross validation, indicating some level of overfitting in case of the two FUV lines.