|
|
Abstract
Decision trees, typically constructed by C4.5, CART or ID3, have been
widely used in machine learning since their conception in the CLS
framework. The single coverage constraint discourages a decision tree to
contain many significant rules. On the other hand, the fragmentation
problem causes a decision tree to contain too many minor rules. Both
shortcomings lead to a reduction in accuracy. Emerging patterns can be
used to solve these problems. Experimental results on gene expression
datasets show that PCL, an EP-based classifier, outperforms single, bagged
and boosted C4.5 trees.
Read the Presentation
Slides...
Referred Papers
|
