Student Exam Performance Prediction

📌 Problem Definition

Predicting student performance on exams is valuable for educators, parents, and students. Understanding the factors affecting student success can guide interventions and improve outcomes.

In this project, we treat it as a machine learning regression problem. Given student characteristics such as gender, ethnicity, parental education level, lunch type, test preparation course, and reading/writing scores, we aim to predict the student’s mathematics score.

This helps answer business questions like:

• Which student features most impact math performance?
• Can schools identify students at risk early?
• How do demographics and study habits correlate with success?

📊 Dataset

We use the Kaggle dataset: “Students Performance in Exams” — Kaggle link

🔍 About the Data

Feature	Type	Description
gender	Categorical	Male or Female
race_ethnicity	Categorical	Ethnicity group (A–E)
parental_level_of_education	Categorical	Parent’s education
lunch	Categorical	Type of lunch received
test_preparation_course	Categorical	Completed or not
reading_score	Numerical	Score in reading exam
writing_score	Numerical	Score in writing exam
math_score	Numerical	Score in mathematics exam (target)

Data characteristics:

• Structured: Tabular features.
• Static: Records collected at one point in time.
• Supervised learning ready: Inputs + target (math_score).

🧪 Evaluation

We measure model performance using:

• Mean Absolute Error (MAE)
• Mean Squared Error (MSE)
• R² Score

Goal: accurately predict math scores. A model with low error and high R² is considered good. For regression, accuracy is about prediction closeness, not a classification percentage.

🔎 Feature Analysis

We use all columns except math_score as features:

Categorical:

• Gender
• Race or ethnicity
• Parental level of education
• Lunch type
• Test preparation course

Numerical:

• Reading score
• Writing score

Categorical features are one-hot encoded. Numerical features are scaled as needed.

🤖 Modelling

Models explored:

• Linear Regression
• Random Forest Regressor
• Gradient Boosting Regressor
• XGBoost Regressor

Modeling Steps:

1. Train-test split
2. Preprocessing: One-hot encode categorical features, scale numerical features
3. Model training
4. Evaluation (MAE, MSE, R²)
5. Model selection based on validation performance

Tree-based models generally performed better than linear regression for this dataset.

🧪 Experimentation

Explored:

• Feature encoding strategies
• Model hyperparameters
• Different regression algorithms

Next steps:

• Cross-validation for robust performance estimation
• Hyperparameter tuning (GridSearchCV / Bayesian optimization)
• Feature importance and explainability (e.g., SHAP)
• Deep learning models (optional)

🚀 Deployment

The final model is deployed as:

• Flask web app — Users enter student details and see predicted math scores.

See application.py

📁 Project Structure

📦 mlproject/
├── .ebextensions/           # AWS Elastic Beanstalk configuration
├── artifacts/               # Serialized model & preprocessor
│   ├── model.pkl
│   └── preprocessor.pkl
├── catboost_info/           # CatBoost metadata (optional)
├── notebook/                # Jupyter notebooks for exploration
├── src/                     # Source code
│   ├── pipeline/
│   │   ├── data_transformation.py
│   │   └── predict_pipeline.py
│   ├── exception.py
│   └── utils.py
├── templates/               # Flask HTML templates
│   └── index.html
├── application.py           # Flask app entry point
├── app_streamlit.py         # Streamlit app entry point
├── requirements.txt         # Dependencies
├── setup.py                 # Optional package setup
├── output.png               # Reference visualization or demo output
└── README.md

📌 How to Run

🐍 Flask App

pip install -r requirements.txt
python application.py

Open in browser:

http://127.0.0.1:5000/

👋 Credits & References

Dataset: Kaggle — Students Performance in Exams

ML concepts: scikit-learn and online resources