Public defence in Computer Science, M.Sc. Arsi Ikäheimonen

Title of the thesis: Advancing Research Methodologies in Digital Phenotyping for Mental Health

Thesis defender: Arsi Ikäheimonen
Opponent: Professor Niels van Berkel, Aalborg University, Denmark
Custos: Associate Professor Mikko Kivelä, Aalto University School of Science

Mental health disorders place a substantial burden on individuals and healthcare systems globally. The assessment and diagnosis of these disorders typically rely on clinical interviews, which can be subjective and require considerable time and clinical resources. To address these challenges, researchers have increasingly used smartphones and other personal devices to unobtrusively collect digital traces of everyday life and behavior. Digital phenotyping is an emerging research field that uses such behavioral data to complement traditional health assessments with objective, real-world information.

This thesis considers digital phenotyping research from two perspectives: methodological and practical. Methodologically, it addresses key challenges that have limited progress in the field, particularly limited reproducibility and comparability across studies. The thesis introduces an openly available, standardized analysis toolbox and outlines a data-driven workflow for behavioral data analysis. These contributions lower the barrier to entry for new researchers by making behavioral analysis more accessible, structured, and transparent.

From a practical perspective, the thesis evaluates the feasibility of smartphone-sensed behavioral data for monitoring and predicting depressive symptom severity in patients diagnosed with depression. The results show that machine-learning models trained on personal smartphone behavioral data can predict changes in depression severity. In addition, the studies demonstrate that collecting such data is feasible in clinical populations, with patients and healthy participants showing similar levels of study engagement and data contribution. The findings further suggest that changes in individual behavior are more informative than differences between individuals, highlighting the importance of personalized prediction modeling and longitudinal monitoring.

Overall, this thesis provides concrete tools, methodological guidelines, and empirical evidence for digital phenotyping research in mental health. Taken together, the findings support future research and the development of complementary, data-driven approaches to understanding and monitoring depression in real-world settings.

Keywords: Digital phenotyping, mental health, depression, smartphone data, machine learning

Thesis available for public display 7 days prior to the defence at Aalto University's public display page.