Predictive Analysis of Mental Health Determinants in Student Populations

Abstract
This research project implements a supervised machine learning framework to identify and classify depression risk factors among students. By analyzing a dataset of over 27,000 observations, the study explores the intersection of academic pressure, socio-economic status, and lifestyle habits to provide a data-driven perspective on student well-being.

Research Objectives

• Risk Classification: Develop high-precision models to categorize individuals based on depressive symptom indicators.
• Feature Criticality: Quantify the impact of variables such as 'Financial Stress', 'Academic Pressure', and 'Study Satisfaction'.
• Lifestyle Correlation: Evaluate the statistical significance of sleep patterns and dietary habits in predicting mental health outcomes.

Dataset Taxonomy The study utilizes a multi-dimensional dataset comprising 18 features across four primary domains:

• Demographics: Age, Gender, City, and Profession.
• Academic Metrics: Degree type, CGPA, Academic Pressure, and Study Satisfaction.
• Lifestyle Factors: Sleep Duration, Dietary Habits, and Daily Study/Work Hours.
• Clinical Indicators: Family History of Mental Illness, Suicidal Thoughts, and Financial Stress levels.

Methodology: Data Preprocessing To ensure the integrity of the statistical inference, a comprehensive preprocessing pipeline was executed:

• Missing Value Management: Systematic imputation or removal of incomplete records to maintain dataset balance.
• Categorical Encoding: Implementation of One-Hot Encoding and Ordinal Mapping to translate qualitative descriptors (e.g., 'Sleep Duration') into computationally viable formats.
• Outlier Mitigation: Statistical filtering of non-physiological values in continuous features (e.g., Age and Study Hours) to reduce model variance.
• Feature Scaling: Standardizing numerical inputs to a uniform scale, preventing magnitude bias during algorithmic training.
• Target Encoding: Harmonizing the 'Depression' label for binary classification tasks.
• Model Validation & Diagnostic: Implementation of Confusion Matrices across all classifiers to provide a granular breakdown of True Positives and False Negatives, ensuring the clinical reliability of the predictions.

Technical Framework

• Language: Python
• Key Libraries: Pandas (Data Manipulation), Scikit-Learn (Machine Learning), Seaborn & Matplotlib (Exploratory Data Analysis).
• Evaluation Metrics: Performance Metrics: While accuracy was monitored, the primary optimization focused on Recall and F1-Score.
  • Recall Optimization: Critical for clinical screening to minimize Type II errors (False Negatives), ensuring that students at risk of depression are correctly identified.
  • Confusion Matrix Analysis: Used as the primary diagnostic tool to evaluate the trade-off between sensitivity and specificity for each implemented model.

Project Deliverables

• Analysis Notebook: Python implementation of the end-to-end ML pipeline.
• Research Presentation: Detailed slide deck summarizing findings and clinical insights.
• Visual Assets: High-resolution infographics and correlation matrices.

Authors

• Isabella Cappiello
• Nathan Dubourg
• Romain Sartori