Overfitting avoidance as bias

Abstract

Strategies for increasing predictive accuracy through selective pruninghave been widely adopted by researchers in decision tree induction. Itis easy to get the impression from research reports that there are statisticalreasons for believing that these overfitting avoidance strategiesdo increase accuracy and that, as a research community, we are makingprogress toward developing powerful, general methods for guardingagainst overfitting in inducing decision trees. In fact, any...