This project is a full-cycle Machine Learning application designed to predict student performance using both linear and polynomial models. Beyond just training models, it showcases modular code design, experiment tracking, MLOps practices, containerization, cloud deployment, and demo hosting.
The goal was simple but ambitious: take raw student performance data and build a scalable, production-ready ML system that can predict outcomes, track experiments, and deploy models in real-world environments.
| Phase | Description |
|---|---|
| 1. Model Development | Trained Linear Regression and Polynomial Regression models. Modularized code into separate scripts, with train.py for training and comparing results against Jupyter Notebooks. |
| 2. Data & Experiment Management | Versioned datasets using DVC, tracked experiments with MLflow for easy comparison of model performance and metrics. |
| 3. API Development | Built a Flask API to serve predictions. Integrated DVC and MLflow into the API to handle data and model versions dynamically. |
| 4. Containerization | Containerized the application using Docker, pushed images to DockerHub. Tested CI/CD workflows with GitHub Actions to ensure builds run smoothly. |
| 5. Cloud Deployment | Pushed Docker images from DockerHub to AWS ECR using IAM roles. Deployed to AWS ECS with cluster, task definition, and service setup. Accessed API via EC2 instances. |
| 6. Model Registry & Demo | Registered models with AWS SageMaker Model Registry for version control. Deployed a demo on Hugging Face Spaces for interactive testing. |
| Layer | Technology |
|---|---|
| ML Models | Linear Regression, Polynomial Regression |
| Experiment Tracking | MLflow |
| Data Versioning | DVC |
| Backend/API | Flask |
| Containerization | Docker |
| CI/CD | GitHub Actions |
| Cloud Deployment | AWS ECR, ECS, EC2 |
| Model Registry | AWS SageMaker |
| Demo Hosting | Hugging Face Spaces |
| Programming Language | Python |
| Visualization / Notebook | Jupyter Notebook |
- Modularized ML pipeline for easy experimentation and reuse
- Versioned datasets and tracked experiments for reproducibility
- Full CI/CD pipeline from local Docker builds to GitHub Actions testing
- AWS Cloud deployment with ECR → ECS → EC2 access
- Demo available publicly via Hugging Face Spaces
- Integration of MLOps tools (MLflow, DVC, SageMaker) for production-ready workflow