- Website launching (23.Jan.2026)
 |  |
Tutorials will take place on Friday the 11th of December 2026. The number of participants to the tutorials is limited and restricted only to those who attend the conference. For further information send an email to info@CMStatistics.org.
Participants will be expected to have their own laptop.
Venue: HTW Berlin, University of Applied Sciences (Wilhelminenhof campus).Title: Flexible Distributional Regression Using the R Package gamlss2
Achim Zeileis, University of Innsbruck, Austria.
Nikolaus Umlauf, University of Innsbruck, Austria.
Email: Contact
Many scientific questions concern not only how the average response changes with covariates, but also how the other properties of the response distribution change, e.g., variability, skewness, tail behavior, quantiles, or exceedance probabilities. Distributional regression addresses these questions by modeling the entire conditional response distribution.
This tutorial introduces generalized additive models for location, scale, and shape (GAMLSS) using the R package "gamlss2" (https://gamlss-dev.github.io/gamlss2), a modern and modular infrastructure for flexible distributional regression. Participants will learn how to choose suitable response distributions, specify additive predictors for multiple distributional parameters, incorporate nonlinear effects and interactions, and interpret covariate effects. Particular emphasis is placed on a complete modeling workflow: Estimation, distributional diagnostics and calibration, model comparison, and probabilistic prediction through conditional quantiles, prediction intervals and exceedance probabilities. The workshop also demonstrates how fitted models can be extended from fast iterative estimation to Bayesian inference using MCMC, enabling posterior uncertainty assessment within the same modeling framework. Selected examples highlight the extensibility of gamlss2 through flexible families, alternative estimation strategies, and special model terms.
Title: Robust prediction and outlier detection
Prof. Christophe Croux , KU Leuven, Belgium.
Email: Contact
The theory and methods of robust statistics are well developed for parametric statistical models. There exist a variety of robust estimators that still give reliable results in presence of model deviation and outliers. The use of a robust estimator allows to detect outliers without suffering from the masking effect. Most of the literature focuses on robust estimation, robust inference, and outlier detection. The prediction part received much less attention.
The first part of the tutorial will review popular robust methods in the regression framework. We also show how they can be used in R. The second part will elaborate more the prediction aspect, including robust methods for time series prediction. Finally, we discuss how machine learning methods can be made more robust. The XGBoost prediction method will be our leading example.