CMStatistics 2022: Start Registration
View Submission - CMStatistics
Title: Nonlinear feature extraction for sufficient dimension reduction in the presence of categorical predictors Authors:  Ben Jones - Cardiff University (United Kingdom)
Andreas Artemiou - Cardiff University (United Kingdom) [presenting]
Abstract: One often encounters regression and classification problems where there is a high-dimensional continuous predictor along with a set of categorical predictors. Existing approaches to nonlinear sufficient dimension regression are unable to accommodate the categorical predictors, while methods that do take them into account seek only linear combinations of the continuous predictors. For the first time, we provide a nonlinear sufficient dimension reduction method which can handle categorical predictors. This is achieved by first extending measure-theoretic developments in sufficient dimension reduction and then adapting a generalised kernel-based version of sliced inverse regression.