Building Behaviour Knowledge Space to Make Classification Decision

Document Type

Conference Proceeding

Publication Date


Find in a Library

Catalog Record


CAEP, namely Classification by Aggregating Emerging Patterns, builds classifiers from Emerging Patterns (EPs). EPs mined from the training data of a class are distinguishing features of the class. To classify a test instance t, the scores by aggregating EPs in t measures the weight we put on each class; direct comparison of scores decides t’s class. However the skewed distribution of EPs among classes and intricate relationship between EPs sometimes make the decision by directly comparing scores unreliable. In this paper, we propose to build Score Behaviour Knowledge Space(SBKS) to record the behaviour of training data on scores; classification decision is drawn from SBKS from a statistical point of view. Extensive experiments on real-world datasets show that SBKS frequently improves CAEP classifiers, especially on datasets where they have relatively poor performance. The improved CAEP classifiers outperform the start-of-the-art decision tree classifier C5.0.


Presented at the 5th Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining, Hong Kong, April 16-18, 2001.



Catalog Record