Well, I model it, although IV is an input that is hard to predict. So no, I don't accept delta as an approximation of probability. Not that it's all that far off, but I don't use it.
How would your method ever predict a negative payout? $ difference is positive, delta difference is positive...