Proportional Membership in Fuzzy Clustering as a Model of Ideal Types

The goal of this paper is to further investigate the extreme behaviour of the proportional membership model (FCPM) in contrast to the central tendency of fuzzy c-means (FCM). A data set from the ¯eld of psychiatry has been used for the experimental study, where the cluster prototypes are indeed extreme, expressing the concept of `ideal type'. While augmenting the original data set with patients bearing less severe syndromes, it is shown that the prototypes found by FCM are changed towards the more moderate characteristics of the data, in contrast with the almost unchanged prototypes found by FCPM, highlighting its suitability to model the concept of `ideal type'.

• Type: Conference
• Year: 2001
• Where: EPIA
• Authors: Susana Nascimento, Boris Mirkin, Fernando Moura-Pires
• Keywords: Artificial Intelligence, Data Set, EPIA 2001, Ideal Type, Proportional Membership Model