Student Performance Predictor

A Flask web app that predicts a student's Math score based on gender, race/ethnicity, parental level of education, lunch type, test preparation course, reading score and writing score. Trained on a labeled student dataset using 9 regression models with the best performer selected automatically via GridSearchCV and tracked using MLflow on DagsHub.

Live: https://data-science-project-1-36kz.onrender.com/predictdata

---

Stack

• Backend: Python, Flask
• ML: Scikit-learn, XGBoost, CatBoost
• Tracking: MLflow, DagsHub
• Deployment: Render with GitHub CI/CD

---

How It Works

User inputs gender, race/ethnicity, parental level of education, lunch type, test preparation course, reading score and writing score through a web form. The request hits a Flask REST API prediction endpoint that loads a preprocessor and trained model from serialized .pkl files and returns the predicted Math score in real time.

Training and prediction pipelines are fully separated — only the prediction pipeline runs on the server.

---

Model Training

9 regression algorithms trained simultaneously:

• Linear Regression, Ridge, Lasso
• Decision Tree, Random Forest
• Gradient Boosting, XGBoost, CatBoost, AdaBoost

Automated hyperparameter tuning using GridSearchCV. Best model selected based on R² score and saved to artifacts/model.pkl.

Best Model: Lasso Regression R² Score: 88.06%

All runs logged via MLflow on DagsHub.

Experiments: https://dagshub.com/kushneek/Data-Science-Project/experiments

---

Data Pipeline

1. Data Ingestion — Reads raw CSV, splits into 80/20 train/test
2. Data Transformation — StandardScaler for numerical features, OneHotEncoding for categorical features, saved as preprocessor.pkl
3. Model Training — GridSearchCV across 9 algorithms, best model saved
4. Prediction — Loads saved artifacts, transforms input, returns prediction

---

Local Setup

1) git clone https://github.com/kushneek/Data-Science-Project.git
cd Data-Science-Project

2) python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

3) pip install -r Requirements.txt
pip install -e .

4) python -m src.mlproject.components.data_ingestion
   python -m src.mlproject.components.data_transformation
   python -m src.mlproject.components.model_training

5) python app.py

App runs at http://localhost:8080/predictdata

---

Project Structure

├── app.py                        # Flask entry point
├── Requirements.txt
├── setup.py
├── src/mlproject/
│   ├── pipeline/
│   │   ├── prediction_pipeline.py
│   │   └── training_pipeline.py
│   ├── components/               # Ingestion, transformation, training
│   ├── logger.py
│   ├── exception.py
│   └── utils.py
└── artifacts/
    ├── model.pkl
    ├── preprocessor.pkl
    └── data/

---

Author:

Kushagra Neekhra — GitHub | LinkedIn