A fuzzy classifier with ellipsoidal regions

Abe, S.; Thawonmas, Ruck

doi:10.1109/91.618273

Cited by 161 publications

(89 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…A logical step was to apply machine learning algorithms to automatically optimise existing systems, or even to create them from scratch using the data. Most of these algorithms were based on supervised learning techniques [1,2,3,13,27,30]. Many of these learning systems (mostly for logic controllers) used fuzzy grids.…”

Section: Precise and Linguistic Fuzzy Modelingmentioning

confidence: 99%

A Fuzzy-XCS Classifier System with Linguistic Hedges

Marín-Blázquez

Shen²

2008

IJCIR

View full text Add to dashboard Cite

Abstract:Many real-world problems require the development and application of algorithms that automatically generate human interpretable knowledge from historical data. Most existing algorithms for rule induction from imprecise data have followed the precise approach, where definitions of the fuzzy sets that are intended to capture certain vague concepts are allowed to be modified such that they fit the data. These approaches typically destroy the original semantics or meaning of the given fuzzy sets, which often leads to loss of transparency in the resulting model or models. In order to overcome this fundamental limitation, a descriptive approach has been proposed in which human defined fuzzy sets are not allowed to be modified. However, as the fuzzy set definitions cannot be modified, and only a small number of them are normally available, only a limited number of possible rules are derivable. Such rules are not very flexible and in many cases, will not necessarily fit the data well. To address this important issue, at least partially, linguistic hedges have been introduced to provide a more adaptable means of learning from data, thereby offering more flexibility in domain knowledge representation and extraction. Following this approach, this paper presents a novel rule induction mechanism which extends a classifier system (XCS) by employing linguistic hedges. The resultant fuzzy XCS classifier with linguistic hedges is evaluated against a real-world forensic glass classification problem. The results demonstrate that the inclusion of hedges to support finer granularity in linguistic fuzzy modelling improves the accuracy of the resulting classifiers, whilst simultaneously preserving the interpretability of the learned models. This approach not only offers the user rules to decide on classes, but also rules to decide which classes to discard. It also inherits from XCS, the ability to deal with data that involves imbalanced classes.

show abstract

Section: Precise and Linguistic Fuzzy Modelingmentioning

confidence: 99%

A Fuzzy-XCS Classifier System with Linguistic Hedges

Marín-Blázquez

Shen²

2008

IJCIR

View full text Add to dashboard Cite

show abstract

“…Such rules have been used in some cases in learning rule systems [20][21][22][23]. However, ellipsoidal surfaces are computationally complex, and all of the above algorithms rely on some clustering method to decide the number and center of the ellipsoids in advance, while the evolutionary algorithm is used for micro-tuning of the parameters.…”

Section: Rule Encodingmentioning

confidence: 99%

Artificial Immune Classifier Based on ELLipsoidal Regions (AICELL) †

Siolas

Stafylopatis

2017

Computation

View full text Add to dashboard Cite

Pattern classification is a central problem in machine learning, with a wide array of applications, and rule-based classifiers are one of the most prominent approaches. Among these classifiers, Incremental Rule Learning algorithms combine the advantages of classic Pittsburg and Michigan approaches, while, on the other hand, classifiers using fuzzy membership functions often result in systems with fewer rules and better generalization ability. To discover an optimal set of rules, learning classifier systems have always relied on bio-inspired models, mainly genetic algorithms. In this paper we propose a classification algorithm based on an efficient bio-inspired approach, Artificial Immune Networks. The proposed algorithm encodes the patterns as antigens, and evolves a set of antibodies, representing fuzzy classification rules of ellipsoidal surface, to cover the problem space. The innate immune mechanisms of affinity maturation and diversity preservation are modified and adapted to the classification context, resulting in a classifier that combines the advantages of both incremental rule learning and fuzzy classifier systems. The algorithm is compared to a number of state-of-the-art rule-based classifiers, as well as Support Vector Machines (SVM), producing very satisfying results, particularly in problems with large number of attributes and classes.

show abstract

“…However, scatter partition of high-dimensional feature spaces is difficult, and thus some learning or automatic evolutionary procedures become necessary [7]. The scatter partition approaches can be further divided into three fuzzy partition methods based on the type of fuzzy regions: hyperbox partition [10], ellipsoid partition [11] and polyhedron partition [12].…”

Section: A Fuzzy Partitionmentioning

confidence: 99%

“…To further realize the performance of the proposed method in the aspects of rule number and test classification rate, we compare the IGA classifier with the existing scatter partition methods: a) fuzzy classifier with hyperbox regions (Hyperbox) [10], b) fuzzy classifier with ellipsoidal regions (Ellipsoidal) [11], and c) neural network classifier with polyhedron regions (Polyhedron) [12]. The test performances of various iris and three rules is (one misclassification).…”

Section: ) Search Ability Of Igamentioning

confidence: 99%

Design of Accurate Classifiers With a Compact Fuzzy-Rule Base Using an Evolutionary Scatter Partition of Feature Space

Chen

2004

IEEE Trans. Syst., Man, Cybern. B

View full text Add to dashboard Cite

Abstract-An evolutionary approach to designing accurate classifiers with a compact fuzzy-rule base using a scatter partition of feature space is proposed, in which all the elements of the fuzzy classifier design problem have been moved in parameters of a complex optimization problem. An intelligent genetic algorithm (IGA) is used to effectively solve the design problem of fuzzy classifiers with many tuning parameters. The merits of the proposed method are threefold: 1) the proposed method has high search ability to efficiently find fuzzy rule-based systems with high fitness values, 2) obtained fuzzy rules have high interpretability, and 3) obtained compact classifiers have high classification accuracy on unseen test patterns. The sensitivity of control parameters of the proposed method is empirically analyzed to show the robustness of the IGA-based method. The performance comparison and statistical analysis of experimental results using ten-fold cross validation show that the IGA-based method without heuristics is efficient in designing accurate and compact fuzzy classifiers using 11 well-known data sets with numerical attribute values.Index Terms-Fuzzy classifier, intelligent genetic algorithm, orthogonal experimental design, scatter partition.

show abstract

A fuzzy classifier with ellipsoidal regions

Cited by 161 publications

References 10 publications

A Fuzzy-XCS Classifier System with Linguistic Hedges

A Fuzzy-XCS Classifier System with Linguistic Hedges

Artificial Immune Classifier Based on ELLipsoidal Regions (AICELL) †

Design of Accurate Classifiers With a Compact Fuzzy-Rule Base Using an Evolutionary Scatter Partition of Feature Space

Contact Info

Product

Resources

About