Bootstrap samples are taken from the original regression or classification data, with the same model selection procedure or classification tree procedure being used on each sample, and the results are aggregated to produce predictions or classifications that have improved precision. See also boosting; machine learning.