Porto Seguro MLOps Project

This repository shows an example of an MLOps pipeline built for the Porto Seguro Safe Driver Prediction dataset. The project demonstrates how to train and deploy a machine learning model using open-source tools and frameworks, including Docker, Kubernetes, and FastAPI. The project serves as an example (and reference) for scalable and production-ready machine learning system design.

Overview

Porto Seguro Dataset

Predicting insurance risk scores is a critical task for reducing losses in the insurance industry. Using the Porto Seguro Safe Driver Prediction dataset, this project develops a machine learning model to predict the likelihood of claims, with an emphasis on high accuracy and operational scalability.

Key Features

• Machine Learning Model:

  • The model is built using XGBoost and optimized using Bayesian Optimization.
  • Performance is measured using the Gini coefficient, a common metric in risk prediction tasks.

• Model Serving:

  • A FastAPI-based REST API serves predictions and provides health checks.

• Scalability:

  • The application is containerized with Docker and deployed using Kubernetes.

Project Structure

porto-seguro-mlops/
├── data/                
├── models/              
│   ├── xgboost_test_gini.pkl
│   ├── model_metadata.json
├── serving/                # FastAPI app for model serving
│   ├── predict.py          # Prediction and health-check endpoints
│   ├── requirements.txt   
├── k8s/                    # Kubernetes manifests
│   ├── deployment.yaml    
│   ├── service.yaml        
├── scripts/                
│   ├── train_model.py      # Script to train the model
│   ├── create_observation.py  
│   ├── run_predict.py      # Script to test the FastAPI app
├── Dockerfile              # Image for teh serving app
├── README.txt              
├── requirements.txt      

Setup Instructions

Prerequisites

Install and set up the following programs:

• Python 3.10+
• Docker (Desktop)
• Kubernetes (Minikube for local testing)
• Git

Step 1: Clone the Repository

git clone https://github.com/<your-username>/porto-seguro-mlops.git
cd porto-seguro-mlops

Step 2: Train the Machine Learning Model

Train and save the machine learning model with metadata:

python scripts/train_model.py

This will generate:

• models/xgboost_gini.pkl: Serialized trained model.
• models/model_metadata.json: Metadata containing feature details.

Step 3: Build and Run the Docker Container Locally

Build the Docker Image:

docker build -t porto-seguro-api .

Tag the Docker Image:

docker tag porto-seguro-api jellewillekes/porto-seguro-api:latest

Replace jellewillekes with Docker Hub username where image is stored.

Push the Docker Image:

docker push jellewillekes/porto-seguro-api:latest

Run the Docker Container Locally:

docker run -p 8080:8080 porto-seguro-api

Test the API:

python scripts/run_predict.py

Using the Deployed Endpoint

The data/observation.csv file contains input data with 91 features. To use the deployed endpoint:

1. Ensure the API is running and accessible. For Kubernetes deployment, retrieve the service URL:

   minikube service porto-seguro-service --url

   Replace <SERVICE_URL> below with the retrieved URL.

2. Send a request to the endpoint:

   curl -X POST -H "Content-Type: application/json" -d '{"features": [<list_of_91_features>]}' <SERVICE_URL>/predict

   Replace <list_of_91_features> with the actual feature values.

3. Alternatively, use the provided script to send requests:

   python scripts/run_predict.py

   This script reads the input data from data/observation.csv, formats it, and sends a request to the endpoint.

The response is a predicted likelihood of an insurance claim based on the provided features.