A prediction space is the space of all possible predictions that a model can produce. For example, a classifier of zoo animals has a prediction space of all potential zoo animals (in concept), while the prediction space of a cat vs dog classifier only contains two animals, cats and dogs.