COVID Student Study: A Year in the Life of College Students during the COVID-19 Pandemic Through the Lens of Mobile Phone Sensing

Title of the Paper

COVID Student Study: A Year in the Life of College Students during the COVID-19 Pandemic Through the Lens of Mobile Phone Sensing

Paper Information

• Field of Study: Research on mobile sensing technology and college students' behavior and mental health
• Keywords: Mobile sensing, COVID-19, pandemic, mental health, digital phenotype

Research Background and Questions

• Background: The COVID-19 pandemic has altered college students' daily lives, including their social interactions, academic activities, and mental health. While many studies have explored behavioral changes during the pandemic, they primarily rely on self-reports, lacking objective data support.
• Identified Issues:
  • The mental health and behavioral changes of college students during the pandemic remain unclear, especially in terms of long-term observations (e.g., the first year of the pandemic).
  • Are COVID-19-related concerns significantly associated with mental health?
  • Can mobile sensing data be used to predict students' mental states (e.g., COVID-19-related concerns) in the future?
• Motivation:
  • Campus pandemic prevention requires an understanding of students' psychological and behavioral changes to formulate targeted policies.
  • Mobile sensing data can reduce the subjective bias of traditional self-reports, providing a more reliable data foundation.

Solution

• Proposed Methods and Solutions:
  • Utilize the "StudentLife" app to capture mobile sensing data (e.g., location, phone usage, sleep patterns) and self-reported questionnaire data from 180 college students, analyzing behavioral and mental health changes during the year prior to the pandemic (March 2019 to February 2020) and the first year of the pandemic.
  • Design a deep learning model (Fully Convolutional Neural Network, FCNN) to predict students' COVID-19-related concerns based on the data.
• Innovations:
  • Combine long-term objective mobile sensing data with behavioral questionnaires to explore broad correlations between college students' mental health and behavior.
  • Use machine learning models to analyze and predict students' mental states, proposing extensive group classification and personalized support mechanisms.
• Implementation Steps:
  1. Data Collection: Continuously collect mobile sensing data and weekly questionnaires from participants over two years.
  2. Data Analysis: Conduct long-term behavioral pattern change analysis, cluster analysis based on self-reported student group classifications, and deep learning predictions of COVID-19 concerns.
  3. Experimental Validation: Evaluate the performance of the prediction model and identify key features influencing COVID-19 concern predictions.

Research Findings

• Specific Findings:
  • Behavioral Change Analysis:
    • During the first year of the pandemic, students' travel distances and activity levels significantly decreased (e.g., walking time reduced by 33%).
    • Average phone usage time increased by 15%, total sleep duration increased by 4%, and time spent at home increased by 38%.
  • Mental Health and COVID-19 Concerns Association:
    • COVID-19 concerns were significantly correlated with depression (ρ=0.35), anxiety (ρ=0.25), and stress (ρ=0.25).
    • 31% of students consistently maintained high levels of concern during the first year of the pandemic.
  • Group Characteristic Differences:
    • Certain student groups exhibited higher levels of depression, anxiety, and stress, along with greater COVID-19 concerns.
    • Significant differences in behavioral changes were observed among groups, such as phone usage habits and sleep patterns.
  • Deep Learning Predictions:
    • FCNN performed best in predicting students' COVID-19 concerns, with an AUROC of 0.70 and an F1 score of 0.71.
    • Key features included unique location visits, time spent at home, and audio playback duration.
• Advantages: Compared to self-reports, mobile sensing data provides continuous, objective, and fine-grained behavioral information, aiding in optimizing interventions and public health policies.
• Limitations and Future Directions:
  • Limitations:
    • The sample is limited to one U.S. university, and the generalizability of the findings needs further validation.
    • Data collection is restricted to smartphones, with some behavioral features dependent on the reliability of underlying APIs.
  • Future Directions:
    • Expand sample diversity (including other countries and different age groups).
    • Introduce multimodal data (e.g., wearable devices) to explore more comprehensive mental health prediction models.
    • Develop digital intervention mechanisms based on prediction models, seamlessly integrating with mental health services.

Conclusion

This study provides critical insights into college students' behavioral and mental health changes during the COVID-19 pandemic. By combining mobile sensing data with deep learning techniques, it not only enhances the understanding of public health crises but also offers data-driven solutions for future policy-making and personalized interventions.