The Improvement of Boosting Algorithm
based on MDL Principle

Boosting (AdaBoost) is an algorithm which improves classifying accuracy of a learning algorithm by making the learning algorithm to generate multiple hypotheses, and finally unifying them. It has already proposed the way which can improve the classifying accuracy more than the old way by setting a confidence value for each hypothesis. Our goal is to improve the classifying accuracy of AdaBoost using C4.5 by setting a confidence value for each decision tree. We decide at which node of a decision tree to calculate a confidence value according to MDL principle is tobe calculated. We also find out the most probable confidence value.