Increasing the number of target classes.

Creating SQL indexes automatically.

Assigning rewards in reinforcement learning.

Reducing dimensionality while preserving as much variance as possible.