Feature rankings from the XGBoost model for HIEROs (red) and other objects (blue). We used the average of the absolute value of the normalised (by predicted redshift) Shapley values. This corresponds to the average percentage impact on model outputs. For example, on average, 8% of the model output for HIEROs is based on the H_E − [4.5] colour, while for other objects the impact of this feature is only around 2%.