Text this: Relative improvements in hotspot prediction accuracy on test sets obtained using models developed on synthetic sampling training sets, compared to those developed on imbalanced training sets.