Yeah, I think it comes from statistics. It's a "model" you "fit" to the data. There is a true (but unknown) function f and you only see noisy Y_i = f(X_i) + eps_i, so f^ is like a model of how f maybe is.