10:10 AM
11:00 AM
Fréchet regression model (Peterson & Müller 2019) provides a promising framework for regression analysis with metric space-valued responses, which are frequently encountered in new statistical applications. However, as in the classical setting, regression accuracy drops significantly as the dimension of the predictor becomes large. We propose a novel sufficient dimension reduction (SDR) framework for Fréchet regression models, which can turn any existing SDR method for Euclidean (X, Y) into one for Euclidean X and metric space-valued Y. Specifically, we map the statistical object to a real-valued random variable by a family of functions, which we call an ensemble, and perform classical SDR on the transformed Y. We show that, when the family of transformations is rich enough, we can assemble the results for transformed responses to fully recover the Fréchet SDR space. The finite sample performance of the methods is illustrated through simulation studies for several special cases that include Wasserstein space, the space of symmetric positive definite matrices and the sphere. The SDR for Fréchet regression can also be used to assist data visualization. This is illustrated for human mortality data.