ABSTRACT
Department of Computing and Information Systems, The University of Melbourne
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 13.2 Emerging Pattern Based Class Membership Score . . . . . . . . . . . . . . 188 13.3 Emerging Pattern Enhanced Weighted/Fuzzy SVM . . . . . . . . . . . . 188
13.3.1 Determining Instance Relevance Weight . . . . . . . . . . . . . . . . 189 13.3.2 Constructing Weighted SVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191 13.3.3 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
13.4 Emerging Pattern Based Weighted Decision Trees . . . . . . . . . . . . . . 193 13.4.1 Determining Class Membership Weight . . . . . . . . . . . . . . . . . 193 13.4.2 Constructing Weighted Decision Trees . . . . . . . . . . . . . . . . . . 194 13.4.3 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 13.4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
13.5 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
This chapter discusses how emerging patterns can be used to help improve traditional classification algorithms. It focuses on two specific approaches, one [140, 461] using emerging patterns (EPs) [118] in weighted/fuzzy support vector machine (SVM) construction, and the other [8] using EPs in weighted decision tree construction. In the first approach, each training data instance is first given an EP based “relevance weight” to reflect its perceived importance for weighted SVMs (three weighting methods are discussed); in the second, each training data instance is first given a “class membership weight vector”, of weighted membership for the classes. As will be seen below, the two approaches lead to significant improvement in classification accuracy and other benefits.
