Title: Semi-supervised classification for functional data and its applications
Authors: Yoshikazu Terada - Osaka University; RIKEN (Japan) [presenting]
Abstract: In various fields, data recorded continuously during a time interval and curve data such as spectral data become common. These kinds of data can be interpreted as ``functional data.'' We have studied binary semi-supervised classification problem for functional data. For example, in the sports medicine field, it is important to identify players who are at-risk for career-threatening injuries based on the various functional data reflecting individual motor dynamics. For this problem, the usual supervised classification methods are not appropriate since it is not necessary that all the at-risk players will have serious injury during the experimental period. In this situation, we consider binary classification problem from only positive and unlabeled functional data. We propose a simple classification algorithm for this problem. In addition, we prove that, under mild regularity conditions similar to those in a supervised context, the proposed algorithm can achieve perfect asymptotic classification in the context of PU classification. In fact, we show that the proposed algorithm works well not only in numerical experiments but also for real data examples.