Authors:
Omid Dehzangi
1
;
Ehsan Younessian
1
and
Fariborz Hosseini Fard
2
Affiliations:
1
Nanyang Technological Universit, Singapore
;
2
SoundBuzz PTE LTD, Singapore
Keyword(s):
Fuzzy Systems, Classification, Iterative Rule Learning (IRL), Rule Weighting, ROC.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Computational Intelligence
;
Decision Support Systems
;
Fuzzy Control
;
Fuzzy Systems
;
Informatics in Control, Automation and Robotics
;
Intelligent Control Systems and Optimization
;
Mobile Robots and Autonomous Systems
;
Robotics and Automation
;
Soft Computing
Abstract:
Fuzzy Rule-Based Classification Systems (FRBCSs) focus on generating a compact rule-base from numerical input data for classification purposes. Iterative Rule Learning (IRL) has been proposed to reduce the search space for learning a rule-set for a specific classification problem. In this approach, a rule-set is constructed by searching for an appropriate fuzzy rule and adding it to the rule-set in each iteration. A major element of this approach is the requirement of an evaluation metric to find the best rule in each iteration. The difficulty in choosing the best rule is that the evaluation metric should be able to measure the degree of cooperation of the candidate rule with the rules found so far. This poses a major difficulty when dealing with fuzzy rules; because unlike crisp rules, each pattern is compatible with a fuzzy rule only to a certain degree. In this paper, the cooperation degree of a candidate rule is divided into the following two components: I)- The cooperation degre
e of the rule with other rules of the same class, II)- The cooperation degree of the rule with rules of the other classes. An IRL scheme to generate fuzzy classification rules is proposed that induces cooperation among the rules of the same class. Cooperation between the rules of different classes is handled using our proposed rule-weighting mechanism. Through a set of experiments on some benchmark data sets from UCI-ML repository, the effectiveness of the proposed scheme is shown.
(More)