Integrated adaptive fuzzy clustering (IAFC) algorithm

Author

Kim, Yong S. ; Mitra, Sunanda

Author_Institution

Dept. of Electr. Eng., Texas Tech. Univ., Lubbock, TX, USA

fYear

1993

fDate

1993

Firstpage

1264

Abstract

An integrated adaptive fuzzy clustering (IAFC) algorithm using a structure similar to that found in the Adaptive Resonance Theory (ART-1) neural network, is presented. The IAFC incorporates a new learning rule and a new similarity measure to eliminate some structural problems inherent in other fuzzy ART-type neural networks. The new learning rule utilizes a fuzzy membership value, a function of the number of iterations, and a fuzzy within-cluster membership value. The new similarity measure incorporates a fuzzy membership value to the Euclidean distance. This incorporation of the new learning rule and the new similarity measure guarantees the convergence of weights in the IAFC algorithm and provides more flexibility to the shapes of the clusters formed by this algorithm. The critical parameters in the operation of IAFC are discussed. The performance of IAFC is evaluated in the classification of real data and compared with other recent neuro-fuzzy clustering algorithms

Keywords

fuzzy logic; iterative methods; knowledge based systems; learning (artificial intelligence); ART-1; ART-type neural networks; Adaptive Resonance Theory; Euclidean distance; convergence of weights; fuzzy membership value; fuzzy within-cluster membership value; integrated adaptive fuzzy clustering; iterations; learning rule; neuro-fuzzy clustering algorithms; similarity measure; Clustering algorithms; Convergence; Euclidean distance; Fuzzy neural networks; Fuzzy sets; Neural networks; Partitioning algorithms; Resonance; Shape measurement; Subspace constraints;

fLanguage

English

Publisher

ieee

Conference_Titel

Fuzzy Systems, 1993., Second IEEE International Conference on

Conference_Location

San Francisco, CA

Print_ISBN

0-7803-0614-7

Type

conf

DOI

10.1109/FUZZY.1993.327574

Filename

327574