This article shows how to deploy a machine learning model as a web API using:
- Flask (to serve the model)
- Docker (to containerize the application)
- Azure App Service (to host it on the cloud)
You are going to be training a model to consuming it live via a curl/Postman request – all in under an hour.
Prerequisites:
- Knowledge of Python – Creating a simple program, run the program from command prompt or bash.
- Docker – Knowledge about basic Docker commands and Docker installed in local.
- Azure – Free or paid subscription and basic knowledge of resource groups, resources, Azure Container Registry and App services.
- Azure CLI installed.
- Bash prompt (Will be using bash scripts for docker and azure).
Overall Project Structure:
breast_cancer_model_deploy/
├── Dockerfile
├── requirements.txt
├── train_model.py
├── serve_model.py
├── breast_cancer_model.pkl ← (generated after running train_model.py)
└── README.md ← (optional, document instructions)
Step 1: Train and Save the Model (Locally)
We are going to use scikit-learn’s Breast Cancer dataset to train a logic regression classifier for this article.
# train_model.py
from sklearn.datasets import load_breast_cancer
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
import pickle
# Load dataset
data = load_breast_cancer()
X_train, X_test, y_train, y_test = train_test_split(data.data, data.target, random_state=42)
# Train model
model = LogisticRegression(max_iter=1000)
model.fit(X_train, y_train)
# Save model to a .pkl file
with open('breast_cancer_model.pkl', 'wb') as f:
pickle.dump(model, f)
print("Model saved to breast_cancer_model.pkl")
# To run this script, use the command:
# python train_model.py
Once you run the program train_model.py, you should see the breast_cancer_model.pkl file created in the program root folder.
breast_cancer_model.pkl – this is the trained model that is going to be used in API code.
Step 2: Serve the Model Using Flask
We are going to write a lightweight API with Flask to expose the model breast_cancer_model.pkl
# serve_model.py
from flask import Flask, request, jsonify
import pickle
import numpy as np
app = Flask(__name__)
# Load model when app starts
with open('breast_cancer_model.pkl', 'rb') as f:
model = pickle.load(f)
# Define the prediction endpoint
@app.route('/predict', methods=['POST'])
def predict():
data = request.get_json()
features = np.array(data['input']).reshape(1, -1)
prediction = model.predict(features)
return jsonify({'prediction': int(prediction[0])}) # 0 = malignant, 1 = benign
if __name__ == '__main__':
app.run(host='0.0.0.0', port=80)
# To run this script, use the command:
# python serve_model.py
Once we run the serve_model.py, an endpoint is exposed through port 80. Endpoint POST /predict – accepts a JSON input and returns the predicted class.
Step 3: Define the dependencies
Create a requirements.txt:
flask==2.2.5
werkzeug==2.2.3
numpy==1.21.2
scikit-learn==0.24.2
This ensures exact versions are installed in Docker for compatibility.
Step 4: Dockerize the Application
Create a Dockerfile:
# Dockerfile
# Base Python image
FROM python:3.8-slim
# Set working directory inside the container
WORKDIR /app
# Install dependencies
COPY requirements.txt requirements.txt
RUN pip install --no-cache-dir -r requirements.txt
# Copy source code
COPY . /app
# Expose port used by Flask
EXPOSE 80
# Start the Flask app
CMD ["python", "serve_model.py"]
This creates a clean, minimal Docker container with your API.
Step 5: Test Locally with Docker
docker build -t breast_cancer_model_image .
docker run -d -p 80:80 breast_cancer_model_image
Send a test prediction:
curl -X POST http://localhost/predict \
-H "Content-Type: application/json" \
-d '{"input": [13.54,14.36,87.46,566.3,0.09779,0.08129,0.06664,0.04781,0.1885,0.05766,0.2699,0.7886,2.058,23.56,0.008462,0.0146,0.02387,0.01315,0.0198,0.0023,14.91,19.31,96.53,678.1,0.1312,0.2776,0.3514,0.152,0.2842,0.07569]}'
Expected output:
{"prediction": 1}
Prediction 1 means benign.
curl -X POST http://localhost/predict \
-H "Content-Type: application/json" \
-d '{"input": [17.99,10.38,122.8,1001.0,0.1184,0.2776,0.3001,0.1471,0.2419,0.07871,1.095,0.9053,8.589,153.4,0.006399,0.04904,0.05373,0.01587,0.03003,0.006193,25.38,17.33,184.6,2019.0,0.1622,0.6656,0.7119,0.2654,0.4601,0.1189]}'
Expected output:
{"prediction": 0}Prediction 0 means malignant.
Step 6: Deploy to Azure App Service
Create Resources
az login
az group create --name bc-rg --location eastus
az acr create --resource-group bc-rg --name bcacr12345 --sku Basic --admin-enabled true
Push Image to ACR
az acr login --name bcacr12345
docker tag breast_cancer_model_image bcacr12345.azurecr.io/breast_cancer_model_image:v1
docker push bcacr12345.azurecr.io/breast_cancer_model_image:v1
Deploy with Azure App Service
az appservice plan create --name bc-plan --resource-group bc-rg --is-linux --sku B1
az webapp create --resource-group bc-rg \
--plan bc-plan --name bc-api-app \
--deployment-container-image-name bcacr12345.azurecr.io/breast_cancer_model_image:v1
Configure Registry Credentials (optional if admin enabled)
az webapp config container set \
--name bc-api-app \
--resource-group bc-rg \
--docker-custom-image-name bcacr12345.azurecr.io/breast_cancer_model_image:v1 \
--docker-registry-server-url https://bcacr12345.azurecr.io
Your ML API is Live!
curl -X POST https://<your-app>.azurewebsites.net/predict \
-H "Content-Type: application/json" \
-d '{"input": [...]}'Final Thoughts
This article walks through a complete ML model deployment pipeline:
- From training to live API
- Containerized and cloud-hosted
- Easily replicable for other models and use cases.
What can you further?
You can use any Dataset instead of one that we used in this article – Here’s how:
- Choose a dataset (e.g. from sklearn, CSV or a database).
- Train any classifier or regressor.
- Save the model with pickle or joblib.
- Update serve_model.py with correct input shape and logic.
- Rebuild the Docker image.
- Expose a /predict endpoint for the new use case.
- Deploy to Azure App Service.
Example use cases you could use
| Use Case | Dataset Example (in sklearn or public) | Target |
| Predict diabetes | sklearn.datasets.load_diabetes() | Continuous value (regression) |
| Predict housing price | sklearn.datasets.fetch_california_housing() | Median house price |
| Predict wine quality/class | sklearn.datasets.load_wine() | Wine class (0, 1, 2) |
| Predict customer churn | Custom CSV with features like tenure, charges, etc. | Custom CSV with features like tenure, charges, etc. |
| Predict handwritten digit | sklearn.datasets.load_digits() | sklearn.datasets.load_digits() |