CHAPTER ONE
1.0 INTRODUCTION
The mental health and wellbeing of students, especially college students, have become a critical
concern globally due to rising rates of psychological distress, anxiety, and depression. The
COVID-19 pandemic has intensified these challenges by disrupting traditional educational
experiences and social interactions, leading to significant impacts on students' mental health.
Understanding and supporting student mental health requires comprehensive, longitudinal data
capturing behavioral, emotional, and environmental factors.
Recent advances in machine learning (ML) and mobile sensing technologies provide powerful
tools to predict and monitor mental health by analyzing rich datasets collected passively through
smartphones and wearable devices, combined with self-reported surveys. Longitudinal studies,
such as the College Experience Study, have tracked student behaviors and mental health over
multiple years, including pre-pandemic, pandemic, and post-pandemic periods, revealing complex
patterns of resilience and vulnerability.
This project aims to develop a predictive mental health and wellbeing support system for students
using machine learning techniques applied to longitudinal behavioral and survey data. By
leveraging passive sensing data, ecological momentary assessments (EMAs), and advanced
interpretable ML models, the system seeks to provide personalized insights and early warnings to
facilitate timely interventions.
The study draws on extensive research including a four-year mobile sensing study at Dartmouth
College that examined mental health dynamics during the COVID-19 pandemic (Nepal et al.,
2024), machine learning approaches to predict mental health problems from annual student health
surveys (Baba & Bunji, 2023), and interpretable hierarchical models for personalized mental
health prediction (Chowdhury et al., 2025). These foundational works highlight the potential of
integrating behavioral data with ML to enhance mental health support for students.
By addressing the challenges of data heterogeneity, model interpretability, and personalization,
this project contributes to the growing field of digital mental health, aiming to improve student
wellbeing outcomes through scalable, data-driven solutions.
1.1 BACKGROUND OF THE STUDY
Recent studies have demonstrated the utility of mobile sensing and machine learning in monitoring
and predicting mental health symptoms among students. Longitudinal studies, such as the College
Experience Study, have tracked behavioral and mental health changes over multiple years,
including pre-pandemic and pandemic periods, revealing significant impacts of COVID-19 on
student wellbeing. Machine learning models, including LightGBM, random forests, and neural
networks, have been applied to health survey data and sensor-derived features to predict mental
health outcomes with promising accuracy. However, challenges remain in model generalizability,
personalization, and interpretability. The integration of behavioral data, ecological momentary
assessments (EMAs), and qualitative insights forms a comprehensive approach to understanding
and supporting student mental health. (Nepal et al., 2024, Baba & Bunji, 2023, Mutawa & Sruthi, 2024)
1.2 JUSTIFICATION OF STUDY
The rising incidence of mental health issues among students, coupled with the limitations
of traditional assessment methods, necessitates innovative solutions for early detection and
intervention. Machine learning-based predictive systems can provide timely, personalized
support by continuously monitoring behavioral indicators and mental health metrics. This
approach addresses gaps in current mental health services, reduces stigma, and enhances
accessibility. The study's focus on interpretable and personalized models ensures that
interventions are tailored to individual needs, improving efficacy and student outcomes.
(Baba & Bunji, 2023, Abdul Rahman et al., 2023)
1.3 AIM AND OBJECTIVES
Aim: To develop a predictive mental health and wellbeing support system for students using
machine learning techniques applied to longitudinal behavioral and survey data.
Objectives:
To analyze longitudinal mobile sensing and survey data to identify behavioral patterns
associated with mental health. (Nepal et al., 2024)
To develop and evaluate machine learning models for predicting mental health status
among students. (Baba & Bunji, 2023)
To create interpretable and personalized prediction frameworks that facilitate targeted
interventions. (Chowdhury et al., 2025)
To assess the impact of the COVID-19 pandemic on student mental health through data-
driven insights. (Mutawa & Sruthi, 2024)
To provide recommendations for integrating predictive systems into student support
services.
1.4 SIGNIFICANCE OF THE STUDY
This study contributes to the advancement of mental health support for students by harnessing
machine learning and mobile sensing technologies. It offers a scalable and data-driven approach
to early detection of mental health risks, enabling proactive interventions. The personalized and
interpretable nature of the models enhances understanding of individual behavioral factors
influencing mental health, supporting tailored care. Additionally, the study provides valuable
insights into the effects of the COVID-19 pandemic on student wellbeing, informing policy and
practice in higher education institutions. (Nepal et al., 2024, Abdul Rahman et al., 2023)
1.5 SCOPE AND LIMITATIONS
Scope:
The study focuses on college and university students, utilizing longitudinal data collected
over multiple years. (Nepal et al., 2024)
It incorporates diverse data sources, including mobile sensing, ecological momentary
assessments, and health surveys. (Baba & Bunji, 2023)
Machine learning models developed emphasize interpretability and personalization.
(Mutawa & Sruthi, 2024)
Limitations:
Data is primarily from specific cohorts and geographic regions, which may limit
generalizability. (Adler et al., 2022)
Mobile sensing data collection faces challenges such as device variability, operating
system updates, and participant compliance. (Chowdhury et al., 2025)
The predictive models depend on the quality and completeness of self-reported survey
data. (Abdul Rahman et al., 2023)
Ethical considerations around privacy and data security are critical and may constrain
data usage. (Mutawa & Sruthi, 2024)