This repository contains a project report on Depression Prediction and Classification using Bayesian Models, conducted as part of the Advanced Bayesian Data Analysis course at TU Dortmund University.
The study explores the application of Bayesian statistics to predict major depressive disorder (MDD) using real-world survey data. Various Bayesian logistic regression models were implemented to analyze relationships between socio-demographic factors and mental health.
- Course: Advanced Bayesian Data Analysis
- Institution: TU Dortmund University
- Technology Used: R, Bayesian Inference, Logistic Regression
- Dataset: Kaggle & Busara Center for Behavioral Economics
- Year: 2024
✔ Explored Bayesian modeling techniques for depression prediction.
✔ Analyzed socio-demographic data to determine risk factors.
✔ Applied multiple Bayesian models, including logistic regression and hierarchical models.
✔ Performed posterior predictive checks, model comparisons, and hypothesis testing.
✔ Used Leave-One-Out Cross-Validation (LOO-CV) for model evaluation.
- Bayesian Logistic Regression: Used to predict depression cases.
- Prior Selection: Defined informative and weakly informative priors.
- Multilevel Models: Incorporated hierarchical structures.
- Model Comparison: Used LOO-CV and AUC analysis.
- Convergence Diagnostics: Assessed model performance and reliability.
- Bayesian logistic regression successfully predicted depression likelihood.
- Age, marital status, and household size were found to impact depression risk.
- Multilevel models provided better predictive performance.
- LOO-CV and AUC metrics helped compare different models.
The full report, including detailed methodology, statistical analysis, and results, is available in this repository.