This is a Pareto optimization with a limit on the size of the formulas. Sure, with a formula of arbitrary size you can fit anything with 100% accuracy, but the task is much harder if what you are looking for are short models.