This research develops a machine learning system that automatically assesses the quality of wearable sensor data before it is used for clinical decision-making. By filtering unreliable signals and distinguishing noise from genuine health events, the approach aims to improve diagnostic accuracy, reduce false alarms, and enable trustworthy medical use of wearable technologies.

This PhD defense presents computational approaches to human behavior across two domains. First, a construct-mining pipeline uses open-ended responses, sentence embeddings, debiasing, HDBSCAN clustering, and intrusion validation to derive a data-driven taxonomy of emotion regulation strategies. Second, mixed-methods and classifiers analyze German Twitter, linking counter-speech styles to hate reduction at scale.

This research examines how beneficial ownership registers can expose corruption in public procurement. By analyzing data quality, corruption adaptation strategies, and ownership complexity, it shows how gaps and missing data can be used to detect fraud and design smarter anti-corruption efforts, ensuring public funds reach essential healthcare.