Machine Learning Implementation for Depression Diagnosis

Overview

This project focuses on analyzing and comparing the performance of Support Vector Machine (SVM), Random Forest, and XGBoost in diagnosing depression based on individual data. As the team leader, I led the implementation of machine learning techniques, ensuring a robust evaluation process that incorporates Non-Parametric Statistical Testing, Feature Engineering, Resampling, and Hyperparameter Tuning. This project was conducted as part of the Computational Intelligence coursework and received a grade of 97/100.

Key Objectives

1. Compare the performance of SVM, Random Forest, and XGBoost using F1-Score.
2. Implement several statistical and model techniques and intepret the results through feature importances and SHAP Analysis.

Techniques

1. Data Preprocessing: Cleaned and transformed individual data for better model training.
2. Exploratory Data Analysis: Visualized each feature based on depression status, including statistical analysis to confirm significance association.
3. Feature Engineering: Reduced the dataframe dimension through PCA.
4. Resampling: Separated the model into several cases: original cleaned dataset, oversampled, undersampled, random sampling.
5. Hyperparameter Tuning: Optimized model performance using Grid Search and Bayesian Optimization.
6. Model Implementation: Trained and evaluated SVM, Random Forest, and XGBoost.