Solving the fragmentation problem of decision trees by discovering boundary emerging patterns

Abstract

The single coverage constraint discourages a decision tree to contain many significant rules. The loss of significant rules leads to a loss in accuracy. On the other hand, the fragmentation problem causes a decision tree to contain too many minor rules. The presence of minor rules decreases the accuracy. We propose to use emerging patterns to solve these problems. In our approach, many globally significant rules can be discovered. Extensive expert. mental results on gene expression datasets show that our approach are more accurate than single C4.5 trees, and are also better than bagged or boosted C4.5 trees.