Figure 2.

Learning curves with $F_1$ score for models trained on the full imbalanced training set (blue) or its balanced subsets (green), using the same test set. The dotted lines show the total max/min score for each training set size. The solid lines (one per subset) represent elementwise median of curves for all CV-repeats. The thick line shows the median score (or elementwise median curve across subsets). The shaded inner area represents the median average deviation (across all CV-repeats) around the median curve(s), and marks a range of the typical performance.