Multi-Mean Gaussian Processes: A probabilistic framework for multi-correlated functional data

Abstract

Like many other organizations, sports federations collected a substantial amount of data these past years. For these federations, a major aim is the detection of high-potential young athletes to integrate top-level structures in anticipation of important upcoming events (especially Paris 2024 Olympic Games). The presentation will focus on data coming from french swimming federation, for which we own all french swimmers performances since 2002 as well as the age it was achieved at. Considering the improvement phenomenon of the sport level as intrinsically functional, we associate a swimmer $i$ with a function $f_i(t)$ that represents its own improvement path. We build this function from the data points through a smoothing procedure with a B-spline basis. We will work on the B-spline coefficients, which are containing the information about improvement paths. Firstly, we use clustering methods on these coefficients to figure out eventual important structures within the swimmers population. Then, performing classification to gather swimmers into level groups will offer a first practical tool helping the young swimmers detection.