MindGrade: Quantifying the Impact of Mental Health on Academic Performance Using Machine Learning

The MindGrade project targets students at the International Islamic University Malaysia, analyzing the relationship between mental health indicators and academic performance (CGPA) using a machine learning pipeline. The study found that approximately one-third of students have mental health issues, but only 6% seek professional help, revealing a significant treatment gap. The project aims to quantify the correlation between mental health burden and academic performance, providing data-driven insights for university welfare policies.

Global higher education is facing a mental health crisis among college students, with issues like depression and anxiety being common. However, there are few quantitative studies on their impact on academic performance in the Southeast Asian academic context. A 2020 survey at the International Islamic University Malaysia showed that about one-third of 101 surveyed students had at least one mental health issue, and only 6% sought professional treatment, indicating a huge treatment gap. Therefore, the MindGrade project built a reproducible machine learning pipeline to quantify the correlation and provide a basis for interventions.

1. Data Processing: Clean raw data (standardize course names, normalize years, fill missing values), construct composite features (mental health burden score = depression + anxiety + panic attacks); 2. Class Imbalance Handling: Use adaptive SMOTE technology to expand the training set;3. Model Comparison: Train and evaluate four classifiers: logistic regression, random forest, XGBoost, and SVM;4. Interpretability: Use SHAP analysis for feature contribution;5. Statistical Testing: Chi-square test for the correlation between mental health features and CGPA.

• Sample Statistics: 101 students aged 18-24, with a highly imbalanced CGPA distribution (90% concentrated in the 3.00-4.00 range);- Mental Health Status: 34.7% have depression, 33.7% anxiety, 32.7% panic attacks, only 5.9% seek treatment;- Model Performance: Logistic regression performed best (accuracy 0.4286, macro-average F1 0.38);- Feature Importance: Academic year and age are the strongest predictors; mental health burden score is more predictive than individual indicators;- Chi-square Test: Due to small sample size and class imbalance, statistical significance was not reached (p-values between 0.06 and 0.72).

1. Prominent Treatment Gap: Only 6% of students seek professional help, requiring institutional intervention;2. Common Comorbidity: Depression, anxiety, and panic attacks are positively correlated; composite burden score is more predictive;3. Model Selection: For small, imbalanced datasets, simple models (logistic regression) outperform complex ones;4. Limitations: Small sample size, cross-sectional data cannot establish causality, self-report bias.

• University Recommendations: Invest in mental health services, eliminate barriers to help-seeking, establish early warning systems;- Future Research: Expand samples to more universities, collect longitudinal data to track dynamic relationships, evaluate intervention effects, develop real-time early warning systems.