Irrelevant Feature And Rule Removal For Structural Associative Classification Using Structure-Preserving Flat Representation

Practical applications of association rule mining often suffer from overwhelming number of rules that are generated, many of which are not interesting or useful for the application in question. Removing irrelevant features and/or rules comprised of irrelevant features can significantly improve the overall performance. Many statistical and constraint based measures are used to discard unnecessary and irrelevant features and rules when vectorial or tabular data is in question. In contrast, the use of such measures is limited in the tree-structured data domain, due to the structural aspects that are not easily incorporated. In this chapter, we explore the use of a feature subset selection measure as well as a number of common statistical interestingness measures via a recently proposed structure-preserving flat representation for tree-structured data such as XML. A feature subset selection is used prior to association rule generation. Once the initial set of rules is obtained, irrelevant rules are determined as those that are comprised of attributes not determined to be statistically significant for the classification task. The experiments are performed using realworld web access trees and property management dataset. The results indicate that where the dataset has more standard structure a large number of insignificant rules will be discarded and accuracy will increase. However, where the tree instances can vary greatly in terms of structure and label distribution among nodes, while many rules are removed and the accuracy increases, there is a significant reduction in coverage rate of the rule set.