CMStatistics 2023: Start Registration
View Submission - CMStatistics
B0288
Title: A new functional data clustering technique based on spectral clustering and downsampling Authors:  Maryam Al Alawi - Sultan Qaboos University (Oman)
Surajit Ray - University of Glasgow (United Kingdom) [presenting]
Mayetri Gupta - University of Glasgow (United Kingdom)
Abstract: A new framework for clustering functional data is presented along with a new paradigm for performing model selection based on downsampling. The clustering framework is a generalisation of the spectral clustering approach and is flexible enough to exploit higher-order features of curves, including derivatives. Extensive comparative studies with existing methods show a clear advantage of the approach over existing functional data analysis clustering approaches. Additionally, a new paradigm is presented for model selection, by introducing the technique of downsampling, which allows the creation of lower-resolution replicates of the observed curves. These replicates can then be used to provide insight into the tuning parameters for the specific clustering techniques. The usefulness of the proposed methods is illustrated through simulations and applications to real-life datasets.