Title: Nonparametric regression and classification with functional, categorical, and mixed covariates
Authors: Jan Gertheiss - Helmut Schmidt University (Germany) [presenting]
Leonie Selk - Helmut-Schmidt-University (Germany)
Abstract: Nonparametric prediction with multiple covariates is considered, in particular categorical or functional predictors (from a Hilbert space, such as the space of square-integrable functions), or a mixture of both. A linear combination of distance measures each calculated on single covariates is proposed, with weights being estimated from the training data. Emphasis is put on the case of a categorical, multi-class response, because the number of corresponding, nonparametric methods found in the literature that can be used with multiple categorical/functional predictors is very limited. The methodology resented is illustrated and evaluated on both artificial and real world data. Particularly it is observed that prediction accuracy can be increased, and irrelevant, noise variables can be identified/removed by ``downgrading'' the corresponding distance measures.