Description

Regression graphs are special chain graphs to represent joint distributions. They are recent versions of graphical models, developed by S. Lauritzen, D.R. Cox, and N. Wermuth. Our purpose is to build such a graph on a multivariate mixed learning data set, containing both discrete and continuous observations (there are some algorithms in R for it). Then we want to make predictions on test data by means of non-parametric regression methods. The research objective is partly to study these kernel-based methods, and partly to develop the models with longitudinal data.

A survey paper on this topic is found on the homepage of Marianna Bolla: https://www.math.bme.hu/~marib/publica/kramli752.pdf. Also see the references therein.

However, you need not study all the models for this research. For interested students, more specific material is available. During the web-course we will consider recent papers on this topic, together with non-parametric regression tools, and use those for prediction that can be the basis of an artificial intelligence, e.g., medical diagnostic systems. Numerical examples are also considered.