koblo4uall/Teenage-Pregnancy-in-Ghana-Using-Explainable-Artificial-Intelligence-Methodology-Pipeline

This repository contains the complete methodology pipeline and implementation for the project "Teenage Pregnancy in Ghana Using Explainable Artificial Intelligence" — an end-to-end machine learning and causal inference study designed to predict, explain, and understand the determinants of teenage pregnancy using demographic and health data. # Teenage Pregnancy in Ghana – Explainable AI Methodology Pipeline # Overview This repository provides the complete **methodology pipeline** for predicting and understanding **teenage pregnancy in Ghana** using advanced **machine learning, explainable AI (XAI), causal inference, and fairness auditing techniques**. The project demonstrates how predictive models can go beyond classification tasks by integrating interpretability, counterfactual reasoning, causal insights, and bias mitigation — all aimed at informing **evidence-based public health interventions**. --- ## Pipeline Structure The codebase is organised into sequential stages, each represented as a notebook or Python module. Together, these stages form a complete analytical workflow: ### 1. Data Collection and Preprocessing - Load secondary demographic and health survey data (e.g., GDHS). - Handle missing values using **MICE (Multiple Imputation by Chained Equations)**. - Encode categorical features using **one-hot** or **ordinal encoding**. ### 2. Data Cleaning and Encoding - Normalize and transform features. - Ensure consistent data types. - Address outliers and scale numeric features where necessary. ### 3. Predictive Base Modeling Framework - Train multiple models: - `RandomForestClassifier` - `XGBoostClassifier` - `LogisticRegression` - `SVC` - Handle class imbalance using **SMOTE**. - Split data into **70% train / 30% test** with stratification. ### 4. Threshold Tuning - Explore alternative probability thresholds (`0.3`, `0.4`, `0.45`, etc.) - Select the optimal threshold based on **recall** and **F1-score** for the minority class. ### 5. Evaluation Metrics - Accuracy - Precision - Recall (Sensitivity) - F1-score - ROC-AUC Curve - Precision–Recall Curve --- ## Explainability and Counterfactual Analysis ### 6. Global Explainability (SHAP) - Compute SHAP values with `TreeExplainer`. - Visualise **feature importance** with bar plots and beeswarm plots. - Identify the most influential predict …