Hybrid objective function of Fuzzy c-Varieties and cross-shape fuzzy cluster extraction
Daisuke Yoshida, Katsuhiro Honda, Akira Notsu, Hidetomo Ichihashi · 2011
This paper proposes an FCM-type clustering method for capturing cross-shape fuzzy clusters in multi-dimensional data spaces. The proposed objective function is a combination of Fuzzy c-Varieties (FCV) and cross-shape cluster extraction in 2-D space, which is an extended linear fuzzy clustering model with local coordinate rotation. FCV is responsible for finding 2-D planes, on which cross-shape prototypes exist. Each prototypical cross is estimated on the FCV prototypes. Fuzzy memberships are updated using a combined clustering criterion of distances between data samples and FCV prototypes and measures for linear clustering on the FCV prototypes with local coordinate rotation.