Predictive utility

Predictive utility#

Machine-learning efficacy (MLE) asks whether generated data train a useful predictor for a declared target. It compares a generated-trained model with a fresh real-trained model on the same real scoring rows.

MLE reports raw AUROC for classification or MAE for regression, their generated-to-real ratio, and a comparison of the models’ predictions. AUROC and its ratio prefer higher values; MAE and its ratio prefer lower values. See the metric guide for examples, target requirements, and repeated model evaluation.