Fdmclust: Functional data model-based clustering using approximation of probability density for a random function in a reproducing Kernel Hilbert space framework
Hanieh Saeidi, Mina Aminghafari, Mehdi Ashkartizabi · Neurocomputing · 2025
This paper proposes a new probability density approximation for functional random variables in the reproducing kernel Hilbert space (RKHS). Based on this approximation, a novel method for model-based clustering of functional data named Fdmclust is introduced. The previous study was based on the Gaussian assumption of the functional data due to the independence used for the covariance kernel. This assumption is not necessarily valid for general cases. The proposed method is applicable to both Gaussian and non-Gaussian functional data, utilizing the projection of functional data onto the Mercer kernel rather than the covariance kernel. To this end, PCA and ICA are used on the obtained projections for Gaussian and non-Gaussian functional random variables, respectively. The estimation of parameters is based on the EM-algorithm. The Fdmclust approach is evaluated on several simulated and real datasets, and the results confirm its efficiency.