Pharmaceutical Statistics Using SAS

2.3. Boosting

Several authors have written thorough summaries of the historical development of boosting (see, e.g., Friedman et al., 2000; Schapire, 2002) For completeness of this work, we provide a brief overview of boosting and include recent findings.

2.3.1. Evolution of boosting

The concept of boosting originated in the machine learning community where Kearns and Valiant (1989) explored weak learning algorithms (algorithms that can classify objects better than random) and strong learning algorithms (algorithms that can classify accurately). The question of the relationship between weak and strong learning algorithms was initially posed, and was addressed soon after by Schapire (1990). In his work, Schapire showed that if a concept was weakly ...