How about

- make your classifier get to 100% accuracy (or close enough).

Also, do it automatically

- use a unsupervised learning stage - then a simple a classifier (so you don't overfit). Repeat, growing the size of your unsupervised, followed by simple classifier, until you get 100%.