Achieving optimal Bayesian classification performance using a novel approach: the 'race to the attractor' neural network model