In order to minimize the overall risk, the category label that can minimize the risk R(c|x) on the sample is
Among them, h∗ is the Bayesian optimal classifier.
