A fuzzy approach to partitioning continuous attributes for classification

Wai Ho Au, Chun Chung Chan, Andrew K C Wong

Research output: Journal article publicationJournal articleAcademic researchpeer-review

48 Citations (Scopus)

Abstract

Classification is an important topic in data mining research. To better handle continuous data, fuzzy sets are used to represent interval events in the domains of continuous attributes, allowing continuous data lying on the interval boundaries to partially belong to multiple intervals. Since the membership functions of fuzzy sets can profoundly affect the performance of the models or rules discovered, the determination of membership functions or fuzzy partitioning is crucial. In this paper, we present a new method to determine the membership functions of fuzzy sets directly from data to maximize the class-attribute interdependence and, hence, improve the classification results. In other words, it forms a fuzzy partition of the input space automatically, using an information-theoretic measure to evaluate the interdependence between the class membership and an attribute as the objective function for fuzzy partitioning. To find the optimum of the measure, it employs fractional programming. To evaluate the effectiveness of the proposed method, several real-world data sets are used in our experiments. The experimental results show that this method outperforms other well-known discretization and fuzzy partitioning approaches.
Original languageEnglish
Pages (from-to)715-719
Number of pages5
JournalIEEE Transactions on Knowledge and Data Engineering
Volume18
Issue number5
DOIs
Publication statusPublished - 1 May 2006

Keywords

  • Classification
  • Data mining
  • Discretization
  • Fuzzy partition
  • Fuzzy sets
  • Membership functions

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Electrical and Electronic Engineering
  • Artificial Intelligence
  • Information Systems

Cite this