|
Digital Library of the
European Council for Modelling and Simulation |
Title: |
Generating Classification Rules From Numerical Data With
Misclassification Cost |
Authors: |
Tomoharu
Nakashima, Yasuyuki Yokota, Gerald Schaefer, Hisao Ishibuchi |
Published in: |
(2006).ECMS
2006 Proceedings edited by: W. Borutzky, A. Orsoni, R. Zobel. European
Council for Modeling and Simulation. doi:10.7148/2006 ISBN:
0-9553018-0-7 20th
European Conference on Modelling and Simulation, Bonn,
May 28-31, 2006 |
Citation
format: |
Nakashima, T., Yokota, Y., Schaefer, G., & Ishibuchi,
H. (2006). Generating
Classification Rules From Numerical Data With Misclassification Cost. ECMS
2006 Proceedings edited by: W. Borutzky, A. Orsoni, R. Zobel
(pp. 79-84). European Council for Modeling and Simulation. doi:10.7148/2006-0079 |
DOI: |
http://dx.doi.org/10.7148/2006-0079 |
Abstract: |
This paper
compares the performance of various rule-based classification systems. In the
classification problems in this paper it is assumed that a misclassification
cost is associated with each training pattern. Thus, the task of
classification is to minimize the total sum of misclassification costs rather
than to maximize the classification rate. If then rules are being generated
from a given set of training patterns. The differences between the
classification systems used in this paper are (a) whether fuzzy sets or
interval sets are used in the antecedent part of if then rules, and (b) how
the consequent part of the if-then rules is determined. In the determination
of the consequent part of if-then rules we consider cost-based and
compatibility-based determination . In cost based determination,
the consequent class of a rule is determined so that the misclassification
costs is minimal over the covered training patterns by the antecedent part of
the rule. On the other hand, in compatibility based determination the
consequent class of an if-then rule is determined from the compatibility of
training patterns covered by the antecedent part of the rule. The grade of
certainty of the rules in both determination types is calculated by using the
compatibility of training patterns from each class. In a series of
computational experiments, we examine the performance of the classification
systems for three real-world pattern classification problems. |
Full
text: |