🩺 Diabetes Predictor

An end-to-end MLOps project that trains and deploys a machine learning model to predict diabetes based on patient health data. Built with FastAPI, Docker, and Kubernetes, this project demonstrates a real-world ML pipeline from training to deployment.

---

🧰 Tech Stack

• Machine Learning: scikit-learn, pandas, joblib
• API Framework: FastAPI
• Containerization: Docker
• Orchestration: Kubernetes
• Dataset: Pima Indians Diabetes Dataset

---

🚀 API Demo

The API provides two main endpoints and interactive documentation powered by Swagger UI.

1. Root Endpoint (/)

A simple health check endpoint to confirm the API is running.

2. Prediction Endpoint (/docs)

The interactive documentation allows you to easily test the /predict endpoint.

3. Example Prediction

You can send the patient's data as a JSON object to the /predict endpoint and receive a boolean prediction.

---

🧠 Machine Learning Model

This project uses a supervised machine learning model to perform a binary classification task (diabetic or not diabetic).

Dataset

The model is trained on the Pima Indians Diabetes Dataset, a standard dataset from the UCI Machine Learning Repository. It contains health data for female patients of Pima Indian heritage.

Features

To keep the model lightweight and fast for an API, we use the following 5 features:

• Pregnancies: Number of times pregnant
• Glucose: Plasma glucose concentration
• BloodPressure: Diastolic blood pressure (mm Hg)
• BMI: Body mass index (weight in kg/(height in m)^2)
• Age: Age (years)

Model

The model is a Random Forest Classifier from scikit-learn.

A Random Forest is an ensemble learning method. Instead of relying on a single decision tree, it builds many individual decision trees during training. When making a prediction, it collects the "vote" from each tree and chooses the class (diabetic or not) that receives the most votes. This approach is highly effective, more accurate than a single tree, and helps prevent overfitting.

---

📁 Project Structure

.
├── app
│   ├── __init__.py             # Makes 'app' a Python package
│   ├── main.py                 # FastAPI app logic
│   ├── model.py                # Model loading and prediction logic
│   ├── schema.py               # Pydantic data models
│   └── training
│       ├── __init__.py         # Makes 'training' a sub-package
│       └── trainer.py          # Model training script
├── .gitignore
├── Dockerfile                  # Multi-stage Dockerfile
├── k8s-deploy.yml              # Kubernetes deployment manifest
├── README.md
└── requirements.txt            # Python dependencies

---

🚀 Local Setup & Running

1️⃣ Clone the Repository

git clone https://github.com/your-username/mlops-diabetes-predictor.git
cd mlops-diabetes-predictor

2️⃣ Create and Activate Virtual Environment

python -m venv .venv
source .venv/bin/activate

3️⃣ Install Dependencies

pip install -r requirements.txt

4️⃣ Train the Model (For Local Testing)

This will run the training script and save diabetes_model.pkl to the project root. The Docker build performs this step automatically.

python -m app.training.trainer

5️⃣ Run the FastAPI App Locally

uvicorn app.main:app --reload

Now access the interactive API docs at 👉 http://127.0.0.1:8000/docs

---

🐳 Docker & ☸️ Kubernetes Deployment

This section explains how to build, push, and deploy the API in the cloud.

1️⃣ Build the Docker Image

The multi-stage Dockerfile handles both model training and packaging.

#Replace 'your-username' with your Docker Hub username
docker build -t your-username/mlops-diabetes-predictor:latest .

2️⃣ Push to Docker Hub

Before deploying to Kubernetes, push the image to your registry.

# Log in (if not already)
docker login

# Push image
docker push your-username/mlops-diabetes-predictor:latest

3️⃣ Deploy to Kubernetes

Edit k8s-deploy.yml to set your correct Docker image: name, then apply:

kubectl apply -f k8s-deploy.yml

4️⃣ Get Your Public Link (URL)

Check the external IP for your LoadBalancer service:

kubectl get service diabetes-predictor-service --watch

Wait until the EXTERNAL-IP changes from <pending> to a real IP (e.g., 20.123.45.67).

Your API will then be publicly available at: http://<YOUR-EXTERNAL-IP>. Access the Swagger UI docs at: http://<YOUR-EXTERNAL-IP>/docs