We propose a fuzzy approach to simultaneously perform clustering and dimensionality reduction of functional data. In particular, functional observations are clustered within a subspace of reduced dimension obtained by constraining the cluster centroids to lie in that subspace. Smoothing is carried out jointly with the clustering procedure and is applied to the cluster centroids in order to control their regularity. The fuzzy partition is obtained by introducing an entropic penalty on the membership distributions, which ensures, under mild conditions, that the estimated membership degrees remain strictly between zero and one. When the tuning parameter associated with the entropic penalty is set to zero, the proposed method reduces to the Regularized Functional Reduced K-Means approach previously introduced in the literature.
A Fuzzy Entropic Approach for Clustering Curves on a Reduced Subspace / Di Mari, F., Rocci, R., Gattone, S.A., Giordano, S.. - (2026), pp. 213-217. (SIS-FENStatS 2026 Roma ) [10.1007/978-3-032-30881-8_35].
A Fuzzy Entropic Approach for Clustering Curves on a Reduced Subspace
Di Mari, Fabrizio;Rocci, Roberto;Giordano, Sabrina
2026
Abstract
We propose a fuzzy approach to simultaneously perform clustering and dimensionality reduction of functional data. In particular, functional observations are clustered within a subspace of reduced dimension obtained by constraining the cluster centroids to lie in that subspace. Smoothing is carried out jointly with the clustering procedure and is applied to the cluster centroids in order to control their regularity. The fuzzy partition is obtained by introducing an entropic penalty on the membership distributions, which ensures, under mild conditions, that the estimated membership degrees remain strictly between zero and one. When the tuning parameter associated with the entropic penalty is set to zero, the proposed method reduces to the Regularized Functional Reduced K-Means approach previously introduced in the literature.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


