jimekersh/DiabetesRisk-NG

Hybrid LSTM + fuzzy inference system for sex-stratified diabetes burden forecasting in Nigeria (1990–2035), with SHAP explainability and a Streamlit app. # DiabetesRisk-NG: A Hybrid Deep Learning and Fuzzy Inference Forecasting System with Built-in Explainability for Sex-Stratified Diabetes Burden in Nigeria, 1990–2035 ## Overview DiabetesRisk-NG is a comprehensive, publication-ready, open-source forecasting system that combines deep learning (LSTM) and fuzzy inference to predict diabetes prevalence in Nigeria from 2023 to 2035. The system provides sex-stratified predictions (Both sexes, Male, Female) and incorporates Nigeria-specific risk factors with built-in explainability using SHAP. ## Key Features - **Hybrid Forecasting**: Combines LSTM neural networks with fuzzy Mamdani inference system - **Sex-Stratified Models**: Separate LSTM models for Both sexes, Male, and Female populations - **Nigeria-Specific Factors**: Incorporates unrecorded alcohol consumption, childhood thinness patterns, and sex-specific risk profiles - **Explainable AI**: SHAP DeepExplainer with force plots and plain-English summaries - **Interactive Web App**: Beautiful multi-page Streamlit application - **Policy Brief Generator**: Automated PDF generation for policy recommendations ## Author **From Jos, Plateau State, Nigeria** ## Project Structure ``` DiabetesRisk-NG/ ├── data/ │ └── noncommunicable_diseases_indicators_nga.csv # WHO GHO NCD indicators ├── assets/ # Auto-created (plots, PDFs, etc.) ├── models/ # Auto-created (trained models) ├── src/ │ ├── __init__.py │ ├── data_prep.py # Data cleaning, pivoting, interpolation, feature engineering │ ├── model.py # LSTM forecasting models │ ├── fuzzy_logic.py # Nigeria-specific fuzzy inference system │ ├── xai.py # SHAP explainability module │ └── utils.py # Utility functions ├── app.py # Main Streamlit application ├── requirements.txt # Python dependencies ├── README.md # This file └── .gitignore # …