Generating Decision Trees with Boughs

Because decision trees use greedy algorithms, as a decision tree grows, lower branches have smaller number of training examples than upper branches so that the reliability of lower branches become worse than the upper branches. Therefore, a single tree may lead to unnecessary tests of attributes due to the data fragmentation. In order to improve the fragmentation problem of decision trees, a decision tree with boughs consisting of exhaustive rules is suggested. The exhaustive rules can be obtained by applying multidimensional association rule algorithms in limited way. An experiment with real world data was performed to see the effect of the method.