B1019
Title: Triclustering algorithm for functional data with a focus on fMRI data
Authors: Jacopo Di Iorio - Penn State University (United States) [presenting]
Nicole Alana Lazar - Penn State University (United States)
Abstract: Triclustering and biclustering have gained significant attention in multivariate data analysis, enabling the identification of coherent subsets in two or three dimensions. While biclustering algorithms for functional data have received considerable attention, the development of triclustering methods specifically tailored for functional data analysis remains relatively unexplored. A novel triclustering method is proposed specifically designed for functional data analysis, leveraging a functional version of the mean squared residue score. The algorithm adopts a divide-and-conquer strategy, demonstrating its efficacy in handling three-dimensional (3D) tensor datasets of functional data. Through the utilization of simulated data, the efficacy of the method is illustrated in uncovering complex patterns and structures in 3D functional datasets. Specifically, the algorithm is applied to functional magnetic resonance imaging (fMRI) data, to show the method's potential at discovering coherent subsets of subjects, ROIs, and time intervals simultaneously, uncovering hidden patterns, correlations, and co-occurrence structures that may not be apparent through traditional clustering approaches.
