Overfitting Avoidance as Bias

摘要: Strategies for increasing predictive accuracy through selective pruning have been widely adopted by researchers in decision tree induction. It is easy to get the impression from research reports that there are statistical reasons believing these overfitting avoidance strategies do increase and that, as a community, we making progress toward developing powerful, general methods guarding against inducing trees. In fact, any strategy amounts form of bias and, such, may degrade performance instead improving it. If often proven successful empirical tests, this due, not methods, but choice test problems. As examples article illustrate, better or worse, only more less appropriate specific application domains. We not—and cannot be—making both powerful general.