Predicting students' happiness from physiology, phone, mobility, and behavioral data.

Area of Science:

• Computational social science
• Affective computing
• Machine learning applications in mental health

Background:

• Student wellbeing is crucial, with happiness linked to mental health outcomes like depression.
• Understanding factors influencing student happiness can inform targeted interventions.
• Previous research has explored digital data for mental health monitoring, but comprehensive models for student happiness are developing.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting undergraduate students' self-reported happiness.
• To identify key behavioral and physiological factors associated with daily fluctuations in student happiness.
• To explore the potential of happiness modeling for early detection of depression risk.

Main Methods:

• Collected multi-modal data from undergraduate students over one month, including physiological signals, location, smartphone logs, and daily surveys.
• Utilized machine learning techniques, including Gaussian Mixture Models and ensemble classification, for predictive modeling.
• Employed feature selection methods to identify significant predictors of happiness.

Main Results:

• Achieved 70% classification accuracy in predicting self-reported happiness on unseen test data.
• Identified specific behavioral factors, such as sleep patterns and social activity, as significant influences on happiness.
• Demonstrated the feasibility of using integrated data sources for robust happiness prediction.

Conclusions:

• Machine learning models can effectively predict student happiness using diverse data streams.
• The study highlights the potential for digital phenotyping in mental health research and early intervention.
• Further research can refine these models to better support student mental wellbeing and identify at-risk individuals.