Student Performance Prediction – End-to-End ML Project

This project predicts students’ math performance based on demographic and academic attributes such as gender, parental education, lunch type, test preparation, and reading/writing scores.
The dataset is sourced from Kaggle – Students Performance in Exams.

---

Objective

To build a machine learning pipeline that can:

• Clean and preprocess the data
• Perform feature engineering and transformation
• Train multiple ML models and optimize hyperparameters
• Save and deploy the best-performing model using a scalable and reusable pipeline

---

Project Structure

ML_Project_1/
│
├── artifacts/                     # Stores serialized models and processed data
│   ├── preprocessor.pkl
│   ├── train.csv
│   ├── test.csv
│   ├── raw.csv
│
├── logs/                          # Application logs
│
├── notebook/                      # Jupyter notebooks for EDA & experimentation
│   ├── 1. EDA STUDENT PERFORMANCE.ipynb
│   ├── 2. MODEL TRAINING.ipynb
│   └── data/data.csv
│
├── src/
│   ├── components/                # Core ML components
│   │   ├── data_ingestion.py
│   │   ├── data_transformation.py
│   │   ├── model_trainer.py             (to be added)
│   │   └── model_hyperparameter_tuning.py (to be added)
│   │
│   ├── pipeline/                  # Training & prediction pipelines
│   │   ├── train_pipeline.py
│   │   ├── predict_pipeline.py
│   │   └── __init__.py
│   │
│   ├── logs/                      # Logging config files
│   ├── exception.py               # Custom exception handling
│   ├── logger.py                  # Logging utilities
│   ├── utils.py                   # Helper functions (e.g., model saving/loading)
│   └── __init__.py
│
├── venv/                          # Virtual environment
│
├── .gitignore
├── requirements.txt               # Required dependencies
├── setup.py                       # Package configuration
└── README.md

---

Tech Stack

• Language: Python 3.10+
• Libraries: numpy, pandas, scikit-learn, matplotlib, seaborn, joblib
• Frameworks: Flask (for deployment)
• Cloud: AWS EC2, Azure Container Instance (planned)
• Version Control: Git + GitHub
• Environment: Virtual Environment / Conda

---

Key Modules

1️⃣ Data Ingestion

• Loads raw data from notebook/data/data.csv
• Splits data into train/test sets
• Saves the processed datasets in artifacts/

2️⃣ Data Transformation

• Handles missing values with SimpleImputer
• Encodes categorical features using OneHotEncoder
• Scales features using StandardScaler
• Saves preprocessor object (preprocessor.pkl)

3️⃣ Model Trainer (upcoming)

• Trains multiple ML models (e.g., Linear Regression, RandomForest, XGBoost)
• Evaluates metrics (R², RMSE, MAE)
• Saves the best-performing model

4️⃣ Hyperparameter Tuning (upcoming)

• Uses GridSearchCV or RandomizedSearchCV for model optimization

5️⃣ Prediction Pipeline (upcoming)

• Loads saved preprocessor and model to predict unseen data

---

Training the Pipeline

# Step 1: Activate environment
venv\Scripts\activate

# Step 2: Install dependencies
pip install -r requirements.txt

# Step 3: Run Data Ingestion
python src/components/data_ingestion.py

# Step 4: Run Data Transformation
python src/components/data_transformation.py

# Step 5: Run Model Training (when implemented)
python src/components/model_trainer.py

---

Deployment (Planned)

AWS EC2

• Containerize the application using Docker
• Deploy the Flask app and trained ML model on an EC2 instance
• Use NGINX or Gunicorn for serving the production app

Azure Container Instance

• Deploy using Azure CLI or the Azure Portal
• Build Docker image and push it to Azure Container Registry (ACR)
• Run and scale the containerized app directly on Azure

---

Results

Model	R² Score	RMSE	MAE
Linear Regression	0.88	5.40	4.22
Lasso	0.83	6.52	5.16

---

Utilities

• Custom Logging: Provides detailed tracking of every step in the ML workflow
• Custom Exception Handling: Ensures robust and clean error management
• Reusable Pipelines: Modularized preprocessing and model training pipelines for flexibility

---

Author

Mayank Meghwal Data Scientist | Machine Learning Engineer

Email: mayankmeg207@gmail.com GitHub: itz-Mayank

---

Future Enhancements

• Implement CI/CD pipeline with GitHub Actions
• Automate deployment using Docker and Kubernetes
• Integrate model monitoring and automated retraining system
• Add support for multi-cloud deployment (AWS + Azure + GCP)

---

License

This project is open-source and available under the MIT License.

---