Modern Approaches for Component-Wise Boosting: Automation, Efficiency, and Distributed Computing With Application to the Medical Domain

MCML Authors

Daniel Schalk

Dr.

* Former Member

→ Group Bernd Bischl
 Statistical Learning and Data Science

Abstract

This thesis focuses on enhancing component-wise boosting (CWB) by improving its efficiency and usability, particularly in high-dimensional feature spaces and distributed data settings. Key contributions include the optimization of the CWB algorithm through Nesterov’s momentum for faster fitting and reduced memory usage, as well as the development of the Autocompboost framework to integrate CWB with AutoML, emphasizing model interpretability. Additionally, the thesis introduces methods for evaluating binary classification models on distributed data using ROC analysis, and presents several R packages (compboost, dsCWB, Autocompboost, dsBinVal) that implement these advances. (Shortened.)

phdthesis Sch23a

---

Dissertation

LMU München. Mar. 2023

---

Authors

D. Schalk

---

Links

DOI

---

Research Area

 A1 | Statistical Foundations & Explainability