Gradient Boosting in Automatic Machine Learning: Feature Selection and Hyperparameter Optimization

MCML Authors

Janek Thomas

Dr.

* Former Member

→ Group Bernd Bischl
 Statistical Learning and Data Science

Abstract

This thesis focuses on automating model selection in AutoML, specifically through gradient boosting techniques like gradient tree and component-wise boosting. It addresses challenges in hyperparameter optimization using Bayesian methods, introduces a new feature selection technique, and proposes an AutoML approach that simplifies the process while maintaining accuracy. Four R packages were developed: mlrMBO for Bayesian optimization, autoxgboost for AutoML, compboost for component-wise boosting, and gamboostLSS for generalized additive models (Shortened.)

phdthesis

---

Dissertation

LMU München. Apr. 2019

---

Authors

J. Thomas

---

Links

DOI

---

Research Area

 A1 | Statistical Foundations & Explainability