ahmedsal95/cvd-risk-prediction-egypt

# CVD Risk Prediction — Egyptian Population A machine learning-based multi-class prediction system for cardiovascular disease (CVD) risk (**Low / Medium / High**) in the Egyptian population, using feedforward artificial neural networks (ANNs) on clinical and biochemical biomarkers. This repository reproduces the methodology of the study *"A Machine Learning-based Prediction System for Cardiovascular Disease for the Egyptian Population"* end-to-end in a single Jupyter notebook: data preprocessing, three ANN architectures, evaluation, and SHAP-based interpretability. ## Overview Cardiovascular disease is the leading cause of death globally and a major burden in Egypt, yet few prediction systems have been tailored to the Egyptian population. This project trains and compares three feedforward ANN architectures — **narrow**, **bi-layered**, and **tri-layered** — to classify patients into CVD risk categories from nine demographic and biochemical features, then uses SHAP to explain which biomarkers drive the predictions. - **Dataset:** 1,000 Egyptian patients (demographic + biochemical records) - **Task:** Multi-class classification — `Heart_Risk` ∈ {Low, Medium, High} - **Models:** Narrow NN (4), Bi-layered NN (8→4), Tri-layered NN (16→8→4) — ReLU hidden layers, Softmax output, Adam optimizer - **Split:** Stratified 70% train / 15% validation / 15% test - **Explainability:** SHAP (SHapley Additive exPlanations) ## Repository structure ``` cvd-risk-prediction-egypt/ ├── data/ │ └── Egypt_Data.xlsx # Raw dataset (1,000 patients, 13 columns) ├── notebooks/ │ └── CVD_Risk_Prediction_Egypt.ipynb # Full, executed, end-to-end pipeline ├── figures/ # All plots exported from the notebook │ ├── 01_eda_class_distribution_and_correlation.png │ ├── 02_model_performance_comparison.png │ ├── 03_confusion_matrices.png │ ├── 04_roc_curves.png │ ├── 05_shap_class_wise_importance.png │ ├── 06_shap_global_feature_importance …