The Promise of Mobile Sensing for Predicting Momentary Emotions

Understanding the intricate dynamics of emotions in daily life is crucial, both from a scientific and clinical perspective. Yet, obtaining a good picture of these emotion dynamics is challenging due to their contextual and person-dependent nature, and because it places a significant burden on participants. Mobile sensing, which involves analysing behavioural and contextual information that is automatically collected by smartphones, offers great potential for revolutionising emotion research by providing continuous and objective data without participant burden. However, the extent to which this kind of data is able to accurately capture momentary emotion dynamics remains unclear. To answer this question, we examine the potential of mobile sensing data to monitor momentary emotions in everyday life through four chapters in the thesis, investigating its benefits, challenges, and limitations.

To get an initial overview of the field, Chapter 1 provides a narrative review of existing studies that employ mobile sensing to monitor emotions. It highlights the unique opportunities offered by mobile sensing and discusses the promise of tracking moment-to-moment and daily average levels of emotions. However, the lack of a standardised corpus of literature and the need to overcome various obstacles indicate that the full potential of mobile sensing for measuring emotions in daily life has yet to be realised.

As a next step, Chapter 2 explores how mobile sensing can be used to predict subjective sleep quality and associated daily mood and depression with data that was collected with a preliminary version of a mobile sensing application. Using a multiverse analysis, various sensors, features, pre-processing options, and statistical prediction models are examined. While the results vary depending on the choices made in the analysis, some models achieve reasonable predictive quality for subjective sleep quality. 

Chapter 3 presents the development and evaluation of m-Path Sense, a novel platform that combines ESM with background mobile sensing capabilities. A pilot study involving 104 participants demonstrates the reliability of data collection through the app. Challenges related to gaps in the data caused by background app limitations are discussed. Despite these challenges, the fusion of ESM and mobile sensing holds promise for digital phenotyping.

In Chapter 4, we finally investigate the robustness of mobile sensing for tracking momentary emotions throughout daily life. Through a multiverse analysis, linear and non-linear models are employed to predict several emotion states. The results reveal that a time model, incorporating basic freely available information, outperforms complex models based on intricate sensing data. The chapter emphasises the importance of cautious optimism when using mobile sensing for emotion detection and to carefully consider choices made during data pre-processing and analysis.