A resampling method that estimates uncertainty by repeatedly training on data with one point removed.