| Bootstrapping provides a method to estimate the quality of a
concept learning algorithm (classifier).
Draw from the data set a new data set with replacement, the sample
being of the same size as the data set. Typically, 1/e = 37% of the
original data set will be unused in the drawn data set.
Train the classifier on the drawn set, and estimate the error on the
unused parts. Averaging the performance on a large number of
experiments gives the final estimate.
|
