PhD Defence Lea Berkemeier | Using Sensors to Monitor and Predict Momentary Feelings - Exploring how multimodal passive sensing of sleep, physical activity, mobility and smartphone use supports within-person predictions of momentary feelings in daily life

Using Sensors to Monitor and Predict Momentary Feelings - Exploring how multimodal passive sensing of sleep, physical activity, mobility and smartphone use supports within-person predictions of momentary feelings in daily life

Lea Berkemeier is a PhD student in the Department of Psychology, Health & Technology. (Co)Promotors are prof.dr. J.E.W.C. van Gemert-Pijnen from the Faculty of Behavioural, Management and Social Sciences, prof.dr.ir. R.M. Verdaasdonk from the Faculty of Science & Technology, dr. H.K.E. Oldenhuis from the Hanze University of Applied Sciences and dr. W. Kamphuis from TNO Soesterberg.

Mental health problems such as stress, depression, and burnout often develop gradually and fluctuate within individuals. Symptoms are typically detected only after they have escalated because assessments partly rely on infrequent retrospective questionnaires that provide static snapshots and are vulnerable to recall bias. Smartphones, wearable sensors, and Ecological Momentary Assessment (EMA) may complement traditional self-reports. Passive sensing can unobtrusively capture behavioral and physiological patterns, while EMA provides high-frequency assessments of momentary feelings, although it may increase participant burden. This dissertation investigates how multimodal passive sensing data, including sleep, physical activity, mobility, and smartphone use relate to momentary feelings and longer-term mental health, and how these data can support personalized monitoring. Two longitudinal cohorts were studied: 16 PhD candidates monitored for nine months in a preventive context and 18 employees returning to work after stress- or burnout-related sick leave monitored for six months in a restorative context. Both studies combined continuous passive sensing, daily EMA, and repeated retrospective questionnaires. Sleep and physical activity consistently emerged as strongest indicators of day-to-day changes in momentary feelings, particularly during return-to-work. Mobility and smartphone use patterns were more heterogeneous, showing predictive value mainly in individualized models. Associations between sensing features and feelings varied substantially across individuals, and subject-dependent machine-learning models consistently outperformed group-level models. Participant's beliefs about behavioral patterns often corresponded with empirically important predictors. Furthermore, EMA and retrospective questionnaires captured complementary aspects of burnout: EMA reflected short-term within-person fluctuations, whereas retrospective measures primarily captured stable between-person differences. Overall, these findings demonstrate the value of personalized, within-person mental health monitoring. Integrating passive sensing with EMA, this dissertation advances person-centered assessments and lays the foundation for personalized monitoring systems that may detect early warning signs and deliver Just-in-Time Adaptive Interventions (JITAIs). Future work should develop hybrid models that gradually personalize predictions as individual data accumulates.