A1239
Title: Detecting shape outliers in functional data via complexity-driven mixture modeling
Authors: Enea Bongiorno - Universita del Piemonte Orientale (Italy) [presenting]
Kwo Lik Lax Chan - Universita degli Studi del Piemonte Orientale (Italy)
Aldo Goia - Universita' del Piemonte Orientale (Italy)
Abstract: The aim is to explore the challenge of identifying atypical shapes within functional data by framing them as the outcome of contamination from high-complexity elements in a mixture model setting. The focus is on three key directions: (i) introducing a complexity measure derived from small ball probabilities, (ii) formulating a mixture model that incorporates this notion of complexity and analyzing its theoretical impact on small ball behavior, and (iii) developing a decomposition algorithm that separates the mixture into interpretable components, thereby facilitating the implicit detection of outliers. The effectiveness of the proposed framework is illustrated through an applied case study.
