Balanced accuracies for different binary classification models trained and tested on SWIRE objects in CDFS. The ‘Labeller’ column states what set of training labels were used to train the classifier, and the ‘Classifier’ column states what classification model was used. ‘CNN’ is a convolutional neural network, ‘LR’ is logistic regression, and ‘RF’ is random forests. Accuracies are evaluated against the expert label set derived from Norris et al. (2006). The standard deviation of balanced accuracies evaluated across the four quadrants of CDFS (Fig. 8) is also shown. The ‘compact’ set refers to SWIRE objects within 1 arcmin of a compact radio component, the ‘resolved’ set refers to SWIRE objects within 1 arcmin of a resolved radio component, and ‘all’ is the union of these sets.