On Grouping and Partitioning Approaches in Interpretable Machine Learning

MCML Authors

Julia Herbinger

Dr.

* Former Member

→ Group Bernd Bischl
 Statistical Learning and Data Science

Abstract

This thesis addresses the challenges of interpreting machine learning models, particularly focusing on the limitations of global explanation methods. It identifies two key issues: the human-incomprehensibility of high-dimensional outputs and the misleading interpretations caused by aggregation bias. The thesis proposes solutions to these problems, such as grouping features for simpler interpretations and using recursive partitioning algorithms to provide regional explanations, ensuring more accurate and understandable insights into model behavior. (Shortened.)

phdthesis Her23

---

Dissertation

LMU München. Dec. 2023

---

Authors

J. Herbinger

---

Links

DOI

---

Research Area

 A1 | Statistical Foundations & Explainability