Models kind of do this through MOS. It's not machine-learning, but they can account for their own biases in the output. If you're looking for something that knows that NAM is always too NW with low center placement in winter, well that's basically what the humans are for, and there's a published catalog of known model biases on one of the national center sites.