The Performance of a Classifier by Testing Only the Significant Events

In ubiquitous environment, too much information exist, and it is not easy to obtain the well classified data from the information. Therefore an algorithm which should be fast and deduce good result is needed. About it, a decision tree algorithm is much useful in the field of data mining or machine learning system for the problem of classification. However sometimes according to several reasons, a decision tree may have leaf nodes which are made of only noise data or include noise data. Therefore it should be excluded from a decision tree. Because those weak leaves is provided wrong results. This paper proposes a method using a classifier, UChoo, for solving a classification problem, and suggests how to exclude weak leaves as foreknow it whether each leaves is weak or not in decision tree. And, the experiment shows a gradient of the performance of a classifier, Uchoo, by changing the threshold for deciding between acceptable leaves and unacceptable leaves.