B0911
Title: Boosting distributional regression models for multivariate responses
Authors: Andreas Mayr - University of Bonn (Germany) [presenting]
Nadja Klein - Georg-August-University Goettingen (Germany)
Abstract: Over the last few years, statistical modelling approaches that go beyond the classical regression of the conditional mean have gained more and more attention. One of the most popular model classes in this context are generalized additive models for location, scale and shape (GAMLSS). The main idea of GAMLSS is that each parameter of the conditional distribution - not only the expected value - is modelled by its own additive predictor. We extend this approach towards multivariate responses and present a statistical boosting algorithm that is able to estimate the unknown quantities of these complex models in potentially high-dimensional settings by circling through the different parameter and outcome dimensions.
