This is an announcement for the next CoMeEcon seminar on variable importance in ML, which will take place on Tuesday, June 9.

• Date & time: June 9, 14.00-17.00
• Location: VU, New University Building, room NU-4B47. A link for online participation will be shared later.

Ángel Reyero Lobo (CWI-Inria-IMT) will present Beyond Coefficients: Understanding Variable Importance in Modern ML.

Abstract

Classical statistical models offer straightforward interpretations through their coefficients, but modern machine learning models often behave as black boxes. This talk explores how variable importance methods evolved to bridge this interpretability gap. After discussing Random Forest importance measures and the emergence of model-agnostic permutation methods, we focus on the challenges posed by correlated variables and conditional dependencies. We present Conditional Feature Importance (CFI) and Leave One Covariate Out (LOCO) as principled alternatives.

Please bring your laptop for the hands-on session.

Further information will be made available at the comeecon web page here

COMEECON is a seminar on Computational statistics, Methodology and Econometrics aimed at early career statisticians. It is part of the Mathematical Statistics section of the VVSOR.