Bayes_XAI

Bayesian logistic regression classifier for cardiovascular risk prediction, built with NumPyro and JAX. Provides per-prediction uncertainty estimates via posterior sampling (NUTS/MCMC).

Stack

• Model — NumPyro (NUTS sampler), JAX
• Preprocessing — scikit-learn pipeline, Pydantic validation
• API — FastAPI + Uvicorn
• Frontend — React + Vite (TypeScript), served via nginx

Project structure

src/
  esitmator/        # BayessianClassifier (fit, predict, predict_proba, predict_uncertainty)
  models/           # NumPyro model definition, PatientRecord Pydantic schema
  preprocessing/    # Preprocessor (StandardScaler + OneHotEncoder pipeline)
  server/           # FastAPI app + Dockerfile
frontend/           # React + Vite frontend + Dockerfile + nginx.conf
data/               # Training data (NSH_clear.csv)
docker-compose.yml  # Compose: server (8000) + frontend (80)
fit_model.py        # Train and evaluate, saves figures/evaluation.png

Dataset

Nigeria Smoking & Health — somtoonkannebe

3 900 patient records from Nigeria with demographic, haematological, and cardiovascular features. Target variable is cardio_risk, derived from blood pressure, cholesterol, and heart rate thresholds.

Setup

Requires uv.

uv sync

Usage

Train and evaluate

uv run python fit_model.py

Saves figures/evaluation.png with confusion matrix, ROC curve, and uncertainty distribution.

Run the server

uv run uvicorn src.server.server:app --host 0.0.0.0 --port 8000

• http://localhost:8000/health — health check (GET)
• http://localhost:8000/predict — REST endpoint (POST)
• http://localhost:8000/docs — Swagger UI

Frontend (dev)

cd frontend && npm install && npm run dev

• http://localhost:5173 — React dev server (proxies /predict to port 8000)

Docker (full stack)

just run    # build and start both services
just down   # stop and remove containers

• http://localhost — frontend (nginx)
• http://localhost:8000 — API

API

POST /predict

{
  "age": 45,
  "sex": 1,
  "cigs_per_day": 0,
  "sickle_cell_genotype": "AA",
  "malaria_exposure": 0.0,
  "hemoglobin_g_per_dL": 13.5,
  "heart_rate_bpm": 78,
  "cholesterol_mg_per_dL": 245.0,
  "blood_pressure_upper": 145.0,
  "blood_pressure_lower": 92.0
}

Response:

{
  "prediction": 1,
  "probability": 0.81,
  "uncertainty": {
    "mean": 0.81,
    "std": 0.09,
    "lower": 0.63,
    "upper": 0.94
  }
}

sex: 0 = female, 1 = male malaria_exposure: 0.0 = rare, 0.5 = recent, 1.0 = chronic