A Comparison of Techniques for Selecting and Combining Class Association Rules

Local pattern discovery, pattern set formation and global modeling may be viewed as three consecutive steps in a global modeling process. As each of these three steps have gained an increased attention in recent years, a great variety of techniques for each step have been proposed, but so far there has been no systematic comparison of the possible choices. In this paper, we survey and evaluate several options for selecting a subset of class association rules and for combining their predictions into a global rule model. Our results confirm that the commonly used Weighted Voting technique is, indeed, a good choice. We can also see that pattern set selection does not seem to have a large impact upon the performance of the rule ensemble.