A0641
Title: Model averaging for global Frechet regression
Authors: Daisuke Kurisu - The University of Tokyo (Japan) [presenting]
Taisuke Otsu - London School of Economics (United Kingdom)
Abstract: The analysis of non-Euclidean complex data is gaining popularity in various domains of data science. In a seminal paper, the concept of regression analysis was generalized to accommodate non-Euclidean response objects. On the other hand, model averaging has a long-standing history in conventional regression analysis and is extensively utilized in the statistical literature. The notion of model averaging to global Frechet regressions is extended, and the optimal property of cross-validation in selecting the averaging weights for minimizing the final prediction error is established. A simulation study demonstrates the excellent out-of-sample predictions achieved by the proposed method.
