Predicting Student Stress and Depression with Machine Learning

This project focused on creating a machine learning system to predict student stress levels and assess PHQ-9 depression scores using psychological, academic, and lifestyle factors. Utilizing Logistic Regression, Random Forest, and XGBoost, the stress prediction task was highly successful, with Logistic Regression achieving the strongest overall accuracy of 90.9%. The analysis highlighted sleep quality, safety, basic needs, and bullying as consistent key predictors of student stress. The subsequent challenge of predicting depression severity was hampered by severe class imbalance. To address this, sophisticated techniques were employed: Stratified K-Fold cross-validation and Bayesian Optimization refined model training, while class imbalance was tackled using SMOTE (Synthetic Minority Oversampling Technique) to generate synthetic samples, and applying class weights to prioritize minority categories. Despite these efforts, depression prediction accuracy plateaued between 56% and 61%, revealing limitations in the existing feature set, with blood pressure identified as the most influential factor in this prediction.

This website!!

This portfolio website is built in a retro 8-bit desktop style, featuring interactive windows that display my projects, skills, and contact information. This is fully hand-coded using HTML, CSS, and JavaScript. The design blends playful nostalgia with modern functionality, making the site both engaging and practical.

I'm currently making other projects. Projects that are finished will be listed