TY  - CHAP
ID  - borgelt2009constraining
T1  - Constraining Shape and Size in Clustering
A1  - Borgelt, Christian
A1  - Kruse, Rudolf
ED  - Okada, Akinori
ED  - Imaizumi, Tadashi
ED  - Bock, Hans-Hermann
ED  - Gaul, Wolfgang
TI  - Cooperation in Classification and Data Analysis
Y1  - 2009
SP  - 13
EP  - 25
PB  - Springer
AD  - Berlin
UR  - http://www.springerlink.com/content/h88n872rj23741n4
M2  - doi: 10.1007/978-3-642-00668-5_2
N2  - Several of the more sophisticated fuzzy clustering algorithms, like the Gustafson-Kessel algorithm and the fuzzy maximum likelihood estimation (FMLE) algorithm, offer the possibility to induce clusters of ellipsoidal shape and differing sizes. The same holds for the expectation maximization (EM) algorithm for a mixture of Gaussian distributions. However, these additional degrees of freedom can reduce the robustness of the algorithms, thus sometimes rendering their application problematic, since results are unstable. In this paper we suggest methods to introduce shape and size constraints that handle this problem effectively.
ER  -