A1236
Title: A functional independent component analysis approach for clustering functional data
Authors: Helena Ortiz - University of Granada (Spain) [presenting]
Christian Acal - University of Granada (Spain)
Francesca Fortuna - G.d'Annunzio University of Chieti-Pescara (Italy)
Alessia Naccarato - University of Roma Tre (Italy)
Ana Maria Aguilera - University of Granada (Spain)
Abstract: A clustering approach for Functional data based on Functional Independent Component Analysis (FICA) is proposed, aimed at identifying group structures that are not always well captured by standard Functional clustering techniques. The method takes advantage of higher-order statistical information to detect non-Gaussian features that are useful for distinguishing between groups. Its performance is evaluated through a comprehensive set of simulation studies, covering different sample sizes, noise levels, and underlying structures, including more challenging cases where group differences are subtle or overlapping. Compared with clustering methods based on spline coefficients or Functional principal components, this approach generally achieves better classification accuracy, especially in more complex scenarios. The results suggest that most of the relevant discriminative information tends to be concentrated in a relatively small number of Independent components, often among the last ones. This fact leads to a compact and informative representation of the Functional data. Overall, these findings support the use of Independent Component representations as a robust and efficient alternative for clustering.
