Critical issues and developments in high-dimensional prediction with biomedical applications

English

Séminaire Probabilités & Statistique

5/04/2012 - 14:00 Anne-Laure Boulesteix (University of Munich) Salle 1 - Tour IRMA

After a brief introduction into prediction methods based on high-dimensional molecular data, I give an overview of two of our recent/ongoing research projects on the assessment of such prediction models. In the first project, we suggest a bias correction method for prediction error estimation when the prediction method is chosen optimally based on the cross-validation results. The second project deals with the assessment and validation of the added predictive value of high-dimensional data compared to clinical data.

The last part of my talk briefly surveys further related research topics, including the combination of pre-processing and validation, the use of random forests in molecular research, and the optimization bias in methodological research.

This is a joint with Christoph Bernau&  Willi Sauerbrei.