Pattern Recognition, 4th Edition

4.22. The Boosting Approach to Combine Classifiers

Boosting is a general approach to improve the performance of a given classifier and is one of the most powerful techniques, together with the support vector machines, that blossomed in the 1990s. Although boosting can be considered an approach to combine classifiers, it is conceptually different from the techniques presented in the previous section, and it deserves a separate treatment. The roots of boosting go back to the original work of Viliant and Kearns [Vali 84, Kear 94], who posed the question whether a “weak” learning algorithm (i.e., one that performs just slightly better than a random guessing) can be boosted into a “strong” algorithm with good error performance. At the heart of a ...