Analyzing the Criteria for Fuzzy Classifier Learning

Lecture Notes in Computer Science, 1997

We examine the principle capabilities and limits of fuzzy classi ers that are based on a nite set of fuzzy if{then rules like they are used for fuzzy controllers, except that the conclusion of a rule speci es a discrete class instead of a (fuzzy) real output value. Our results show that in the two{dimensional case, for classi cation problems whose solutions can only be solved approximately by crisp classi cation rules, very simple fuzzy rules provide an exact solution. However, in the multi{dimensional case, even for linear separable problems, max{min rules are not su cient.

International Journal of Approximate Reasoning, 1996

In our preuious work we deueloped a method for extracting fuzzy rules directly from numerical data for pattern classification. The performance of the fuzzy classifier det ,eloped using this methodology was comparable to the auerage per~brmance of neural networks. In this paper, we further deuelop two methods, a least squares method and an iteratit'e method, for tuning the sensitiuity parameters of fuzzy membership functions by which the generalization ability of the classifier is improued. We eualuate our methods using the Fisher iris data and data for numeral recognition of uehicle license plates. The results show that when the tuned sensitiuity parameters are applied, the recognition rates are" improued to the extent that performance is comparable to or better than the maximum perJbrmance obtained by neural networks, but with shorter computational time.

Fuzzy Sets and Systems, 2007

Many image processing applications involve a pattern classification stage. In this paper we propose a classifier based on fuzzy if-then rules that allows the incorporation of weighted training patterns which can be used to adjust the sensitivity of the classification with respect to certain classes. The antecedent part of fuzzy if-then rules are specified by partitioning each attributes into fuzzy sets while the consequent class and the degree of certainty are determined from the compatibility and weights of training patterns. We also introduce a learning method which adjusts the degree of certainty in order to provide improved classification performance and reduced costs. Experimental results on several image processing tasks demonstrate the efficacy of the proposed method.

1999

We present a class of Learning Classifier Systems that learn fuzzy rule-based models, instead of interval-based or Boolean models. We discuss some motivations to consider Learning Fuzzy Classifier Systems (LFCS) as a promising approach to learn mappings from real-valued input to real-valued output, basing on data interpretation implemented by fuzzy sets. We describe some of the approaches explicitly or implicitly referring to this research area, presented in literature since the beginning of the last decade. We also show how the general LFCS model can be considered as a framework for a wide range of systems, each implementing in a different way the modules composing the basic architecture. We also mention some of the applications of LFCS presented in literature, which show the potentialities of this type of systems. Finally, we introduce a general methodology to extend reinforcement distribution algorithms usually not designed to learn fuzzy models. This opens new application possibilities.

Fuzzy Sets and Systems, 2007

This paper considers the automatic design of fuzzy rule-based classification systems from labeled data. The classification accuracy and interpretability of generated rules are of major importance in fuzzy classification systems. We propose a weighting function for compatibility grade of patterns that improves the performance of fuzzy classification system without degrading the interpretability of fuzzy rules. Our approach divides the covering subspace of each fuzzy rule into two subdivisions based on a threshold. Any pattern with compatibility grade above this threshold should be classified truly so the weighting function enhances their association degree. For patterns below threshold, their compatibility grades remain unchanged. The splitting threshold for each rule (i.e. the compatibility grade of a specific pattern) is found using distribution of patterns in the covering subspace of that rule. We also show that how the proposed approach is applicable when fuzzy rules have certainty grades. Experiments on some well-known data sets are used to evaluate the performance of our approach.

IEEE International Conference on Fuzzy Systems, 1993

The authors propose a general fuzzy classification scheme with learning ability using an adaptive network. System parameters, such as the membership functions defined for each feature and the parameterized t-norms used to combine conjunctive conditions, are calibrated with backpropagation. To explain this approach, the concept of adaptive networks is introduced and a supervised learning procedure based on a gradient descent algorithm is derived to update the parameters in an adaptive network. The proposed architecture is applied to two problems: two-spiral classification and Iris categorization. From the experimental results, it is concluded that the adaptively adjusted classifier performs well on an Iris classification problem. The results are discussed from the viewpoint of feature selection

5th Online World Conference on Soft Computing in Industrial Applications, pp. 4-18, 2000

Data-based identification of fuzzy rule-based classi-fiers for high-dimensional problems is addressed. A crisp binary decision tree approach is proposed for the selection of the relevant features and effective initial par-titioning of the input domain. The decision tree is then transformed into a fuzzy rule-based classifier. Fuzzy classifiers have more flexible decision boundaries than decision trees and can therefore be more parsimonious. Therefore, the DT initialized fuzzy classifier is reduced in an iterative scheme by means of similarity-based rule-reduction. A genetic algorithm with a multi-objective criterion searching for both redundancy and accuracy is applied. The proposed approach is studied on an arti-ficial problem and the Wisconsin Breast Cancer classifi-cation problem.

2004

In this paper it is pointed out that a classification is always made taking into account all the available classes, ie, by means of a classification system. The approach presented in this paper generalizes the classical definition of fuzzy partition as defined by Ruspini, which is now conceived as a quite often desirable objective that can be usually obtained only after a long learning process. In addition, our model allows the evaluation of the resulting classification, according to several indexes related to covering, relevance and overlapping.

2004

Designing classifiers may follow different goals. Which goal to prefer among others depends on the given cost situation and the class distribution. For example, a classifier designed for best accuracy in terms of misclassifications may fail when the cost of misclassification of one class is much higher than that of the other. This paper presents a decision-theoretic extension to make fuzzy rule generation cost-sensitive. Furthermore, it will be shown how interpretability aspects and the costs of feature acquisition can be accounted for during classifier design. Natural language text is used to explain the generated fuzzy rules and their design process.

ECMS 2011 Proceedings edited by: T. Burczynski, J. Kolodziej, A. Byrski, M. Carvalho, 2011

In this paper we first introduce the concept of classification confidence in fuzzy rule-based classification. Classification confidence shows the strength of classification for an unseen pattern. Low classification confidence for an unseen pattern means that the classification of that pattern is not very clear compared to that with high classification confidence. Then we focus on the minimum classification confidence for fuzzy rule-based classifiers using the classification confidence. The minimum classification confidence represents the worst classification among given training patterns. Some discussion on assigning a weight to training pattern is given to show that cost-sensitive fuzzy rule-based classifiers are advantageous for producing a large minimum-confidence classifiers. A series of experiments are done in order to show that reasonable classification boundaries can be obtained by cost-sensitive fuzzy rule-based classifiers if appropriate weights are assigned to training patterns.