🎓 Student-Marks-Predictor

A production-ready machine learning web application that predicts student math scores using demographic and academic features, deployed on AWS with a fully automated CI/CD pipeline.

---

📌 Project Overview

This project demonstrates how machine learning models can be deployed into production using Flask, AWS Elastic Beanstalk, and automated CI/CD pipelines. The system processes structured student data, applies preprocessing transformations, and predicts math scores using a trained regression model. The application is lightweight, scalable, and built with a clean modular ML architecture suitable for portfolio and real-world deployment scenarios.

---

✨ Features

🎯 Regression-Based Score Prediction

Predicts Math Score based on:

• Gender
• Race/Ethnicity
• Parental Level of Education
• Lunch Type
• Test Preparation Course
• Writing Score
• Reading Score

---

🧹 Data Processing & Feature Engineering

• Handles missing values
• Applies:
  • One-Hot Encoding for categorical variables
  • Standard Scaling for numerical variables
• Uses a reusable preprocessing pipeline
• Saves preprocessor object for inference reuse

---

🧠 Model Training & Evaluation

• Evaluated multiple regression models
• Selected best model based on R² Score
• Saved trained model using pickle
• Separated training and inference pipelines

---

💾 Model & Artifact Serialization

• Stores trained model (model.pkl)
• Stores preprocessing pipeline (preprocessor.pkl)
• Eliminates retraining during deployment
• Optimized for inference-only production usage

---

🌐 Web Application (Flask)

• Clean and minimal UI
• User-friendly form input
• Displays predicted math score
• Hosted on AWS Elastic Beanstalk
• Powered by Gunicorn WSGI server

---

☁ Fully Automated CI/CD Pipeline

• GitHub → AWS CodePipeline → CodeBuild → Elastic Beanstalk
• Automatic deployment on every push
• No manual zip uploads required
• Versioned deployments via Elastic Beanstalk

---

⚡ Lightweight & Cloud-Optimized

• Optimized for AWS t3.micro instance
• Removed heavy experimental libraries
• Training separated from production runtime
• Fast startup and low memory usage

---

🧩 Modular & Scalable Codebase

The project is built in clearly separated stages:

1. Data Ingestion
2. Data Transformation
3. Model Training
4. Prediction Pipeline
5. Web App Deployment
6. CI/CD Automation

---

📁 Project Structure

• .ebextensions/
  • python.config
• Notebook/
  • Dataset/
    • Student.csv
  • catboost_info/
    • learn
    • catboost_training.json
    • learn_error.tsv
    • time_left.tsv
  • EDA Student Performance.ipynb
  • Model Training.ipynb
• artifacts/
  • data.csv
  • model.pkl
  • preprocessor.pkl
  • test.csv
  • train.csv
• src/
  • components/
    • init.py
    • data_ingestion.py
    • data_transformation.py
    • model_trainer.py
  • Pipeline/
    • init.py
    • predict_pipeline.py
    • train_pipeline.py
  • init.py
  • exception.py
  • logger.py
  • utils.py
• templates/
  • home.html
  • index.html
• .gitignore
• Procfile
• README.md
• app.py
• application.py
• buildspec.yml
• requirements.txt
• setup.py

---

📊 Model Performance

• Evaluation Metric: R² Score
• Final Selected Model: Linear Regression
• Designed for structured tabular data prediction

---

🛠️ Tools & Technologies Used

• Python 3.11
• Flask
• Scikit-learn
• Pandas & NumPy
• Gunicorn
• AWS Elastic Beanstalk
• AWS CodePipeline
• AWS CodeBuild
• GitHub

---

▶️ How to Run the Project Locally

pip install -r requirements.txt python app.py

---

▶️ How to Run the Project through URL

http://student-marks-predictor.us-east-1.elasticbeanstalk.com

---

👤 Author

• Prakhar Srivastava
• Aspiring Data Scientist & Business Analyst | Machine Learning, Deep Learning & Generative AI Enthusiast