Title :
On Fuzzy Clustering Based Self-Organized Methods
Author :
Sato-Ilic, Mika ; Kuwata, Tomoyuki
Author_Institution :
Fac. of Syst. & Inf. Eng., Tsukuba Univ.
Abstract :
This paper presents two methods based on self-organized dissimilarity. The first is an implemented fuzzy clustering and the second is a hybrid method of fuzzy clustering and multidimensional scaling (MDS). Specifically, a self-organized dissimilarity is defined that uses the result of fuzzy clustering in such a way that the dissimilarity of objects is influenced by the dissimilarity of the classification situations corresponding to the objects. In other words, the dissimilarity is defined under an assumption that similar objects have similar classification structures. Through empirical evaluation the proportion and the fitness of the results of the method, which uses MDS combined with fuzzy clustering, is shown to be effective in real data. Furthermore, by exploiting the self-organized similarity, defuzzification of fuzzy clustering can cope with the inherent classification structures
Keywords :
data analysis; fuzzy set theory; pattern clustering; self-organising feature maps; classification structures; defuzzification; fuzzy clustering; multidimensional scaling; object dissimilarity; self-organized dissimilarity; self-organized methods; Clustering methods; Data analysis; Extraterrestrial measurements; Fuzzy logic; Multidimensional systems; Neural networks; Optimization methods; Systems engineering and theory;
Conference_Titel :
Fuzzy Systems, 2005. FUZZ '05. The 14th IEEE International Conference on
Conference_Location :
Reno, NV
Print_ISBN :
0-7803-9159-4
DOI :
10.1109/FUZZY.2005.1452526