Reducing on the number of testing items in the branches of decision trees

Abstract

Even though decision trees are one of the mostly used data mining methods, the cost of testing long branches is a hindrance for usability of the method, if some feature values in the branches require high costs to get or are not available. As a method to overcome this difficulty, we applied a multidimensional association rule algorithm with some restriction to the branches of generated decision tree, and found that most of the branches in the decision tree have shorter and more reliable multidimensional association rules as subsets of the branches so that reducing the number of testing items may be possible. Therefore, by using the found shorter and reliable rules, costs related to testing each item in the branches can be reduced. © Springer-Verlag Berlin Heidelberg 2004.