End to End Machine Learning Project

---

From Model to Production: Dual AWS Deployment (Elastic Beanstalk + Docker/ECR/EC2 CI/CD)

Student Math Score Predictor

This repository is the Elastic Beanstalk deployment track on AWS for the Student Math Score Predictor project.
The Docker + ECR + EC2 CI/CD track is maintained in a separate companion repository.

Repository Map

1. Elastic Beanstalk deployment (this repo): https://github.com/Affan005-ai/ML-Projects
2. Docker + ECR + EC2 CI/CD repo: https://github.com/Affan005-ai/AWS-CI-CD-Projects

Problem Statement

Predict a student’s math score from:

1. Gender
2. Race/Ethnicity
3. Parental level of education
4. Lunch type
5. Test preparation course
6. Reading score
7. Writing score

Project Features

1. End-to-end ML pipeline from ingestion to artifact generation
2. Flask web app for live prediction
3. Multi-model regression training
4. Hyperparameter tuning with RandomizedSearchCV
5. AWS Elastic Beanstalk deployment setup
6. Real commit-driven iteration and fixes

End-to-End Architecture

1. Data ingestion reads Notebook/data/StudentsPerformance.csv
2. Train-test split saved to artifacts/train.csv and artifacts/test.csv
3. Transformation pipeline handles imputation, encoding, and scaling
4. Multiple regressors are trained and tuned
5. Best model + preprocessor saved in artifacts/
6. Flask app serves inference via /predict
7. Elastic Beanstalk hosts the application

ML Pipeline (Deep Dive)

1) Data Ingestion

Implemented in src/components/data_ingestion.py:

1. Reads source dataset
2. Saves raw/train/test CSV artifacts
3. Uses deterministic split with random_state=42

2) Data Transformation

Implemented in src/components/data_transformation.py:

1. Numerical features: impute median + standard scaling
2. Ordinal features: impute most frequent + ordered encoding
3. Nominal features: impute most frequent + one-hot encoding

3) Model Training and Selection

Implemented in src/components/model_trainer.py:

1. Trains multiple regressors
2. Compares test R2 scores
3. Selects best model above quality threshold

4) Hyperparameter Tuning

Implemented in src/utils.py via evaluate_models(...):

1. Uses RandomizedSearchCV
2. n_iter=9, cv=3, scoring='r2', n_jobs=-1
3. Picks best estimator per model
4. Saves best overall model artifact

Inference Layer

application.py exposes:

1. / for UI
2. /predict for form-based prediction

Artifacts loaded at runtime:

1. artifacts/model_1.pkl
2. artifacts/preprocessor_1.pkl

Deployment Track (Primary in This Repo): Elastic Beanstalk

Key files:

1. .elasticbeanstalk/config.yml
2. .ebextensions/python.config
3. Procfile
4. application.py

Why this track:

1. Managed platform lifecycle
2. Faster app hosting with less infra overhead

Deployment Track (Related Repo): Docker + ECR + EC2 + GitHub Actions

Companion repo: https://github.com/Affan005-ai/AWS-CI-CD-Projects

What that track covers:

1. Docker image build
2. Push to Amazon ECR
3. Deploy on EC2 self-hosted runner
4. CI/CD automation in GitHub Actions

Production-Grade Improvements

1. Use Gunicorn-based runtime consistently
2. Add CI quality gates (real tests/linting)
3. Add image vulnerability scanning
4. Use OIDC/IAM role instead of long-lived keys
5. Add CloudWatch logging and alarms
6. Add health checks and rollback strategy

One Screenshot Proof

Quick Start

pip install -r requirements.txt
python application.py
Open:

http://127.0.0.1:5000

Credits

• This project and deployment learning journey were inspired by Krish Naik and his end-to-end ML engineering guidance.

Acknowledgment

• Thanks to the open-source Python, scikit-learn, Flask, Docker, and AWS communities for documentation and tooling support.