amanbelays/HIV_status_Prediction

This repository contains the code and documentation for the internship project titled "Building Machine Learning Models to Predict HIV Status Using the Ethiopian HIV Impact Survey Data." The repository includes data preprocessing scripts, model selection and training code, evaluation metrics, and visualization techniques. # Predicting HIV Status Using Machine Learning This repository contains my internship project at the Ethiopian Public Health Institute (EPHI) - National Data Management Center (NDMC). The project explores machine learning (ML) approaches to predict HIV status based on demographic, behavioural, and socioeconomic factors from the Ethiopian HIV Impact Assessment (EPHIA) survey. ## 🔍 Project Overview The goal of this project is to leverage ML techniques to improve HIV status prediction accuracy, providing valuable insights into Ethiopia’s HIV epidemic, particularly in urban areas. ### 📌 Key Objectives - Utilize ML models to predict HIV status. - Handle challenges such as imbalanced data and missing values. - Compare different ML models and optimize performance. - Provide insights into the public health implications of HIV prediction. ## 📊 Dataset - **Source:** Ethiopian Population-based HIV Impact Assessment (EPHIA) (2017-2018) - **Size:** 11,810 observations and 312 variables - **Features Used:** 11 independent variables selected through literature review - **Target Variable:** HIV status (positive/negative) ## ⚙️ Methods & Tools - **Data Preprocessing:** Handling missing values, feature engineering, standardization - **Machine Learning Models:** - Logistic Regression - Decision Trees - Random Forest (Final Model) - Gradient Boosting - **Techniques Applied:** - SMOTE for handling imbalanced data - Hyperparameter tuning (cross-validation) - Performance evaluation (AUC-ROC, F1-score, accuracy) - **Libraries Used:** -'tidyverse', 'skimr', 'GGally', 'tidymodels','usemodels', 'VIP, 'themis' ## 📈 Results - The **Random Forest model** demonstrated the best performance. - Evaluation metrics: - Accuracy: **96.5%** - ROC-AUC: **0.78** - Sensitivity: **99.1%** - Specificity: **6.2%** - Feature importance analysis revealed the following key predictors of HIV status. - ## 📌 Challenges Faced - Limited domain expertise required extensive literature review. - High-dimensional da …