Flow diagram summarising the main steps in our chained regression implementation. In the first step, a CatBoostRegressor model is trained using the training data features X and training data labels y (not shown) for one of the galaxy properties as inputs. The resulting model then provides predictions $\widehat{y_{p,i}}$ for this galaxy property, both for the test set and the training set. These predictions are merged into to the training and test datasets as a new feature. This process is continued until each property has been predicted the required number of times, at which point the loop is stopped and the final predictions for each property are obtained.