A Multivariate Copula-Based Conformal Bayesian Framework for Doping Detection

Nina Deliu (Sapienza University of Rome)

Friday 9th May 11:00-12:00 Maths 311B

Abstract

This work addresses the problem of detecting abnormal values in multivariate longitudinal data using conformal inference. Specifically, motivated by the mission of the world anti-doping agency (WADA), the interest is in identifying potential doping abuse in sports by analyzing athletes’ individual profiles over time. We propose a solution based on conformal predictive inference within a multivariate Bayesian hierarchical framework. To facilitate tractable and accurate estimation of multivariate profiles, we leverage the use of copula models, which allow for separate modeling of the marginal distributions and their dependence structure. Bayesian principles are employed to estimate the multivariate predictive distribution, and the correspondent multidimensional reference region is derived using conformal theory. The practical implementation and relevance of this framework is demonstrated in simulations and in real profile data on a control (non-doped) population and on a doped athlete. Based on joint work with Brunero Liseo.

Add to your calendar

Download event information as iCalendar file (only this event)