Decision functions from supervised machine learning algorithms as collective variables for accelerating molecular simulations