Home | Publications | Sim26

Re-Evaluating the Machine Learning Pipeline to Improve Fairness and Reliability

MCML Authors

Jan Simson

→ Group Christoph Kern
Social Data Science and AI
→ Co-Group Frauke Kreuter

Abstract

This dissertation critically examines the machine learning pipeline in algorithmic decision-making systems, highlighting how numerous design choices can influence fairness, bias, and system outcomes. Using the concept of multiverse analysis, it systematically studies alternative decision paths, proposes methods to improve robustness and transparency in fairness research, and introduces tools and datasets to support more responsible ML development. (Shortened.)

phdthesis Sim26