Insurance Premium Predictor

A machine learning solution for predicting insurance premiums using LightGBM. This project demonstrates end-to-end ML pipeline development with focus on code quality and maintainability.

🎯 Project Overview

This project implements a robust machine learning pipeline for predicting insurance premiums. It features:

• Custom feature engineering for insurance data
• LightGBM model with optimized parameters
• 5-fold cross-validation
• RMSLE (Root Mean Squared Logarithmic Error) optimization
• Type-safe implementation with comprehensive error handling

🛠️ Technical Stack

• Python: 3.11+
• Core Libraries:
  • lightgbm: Gradient boosting framework
  • pandas: Data manipulation
  • numpy: Numerical operations
  • scikit-learn: ML utilities

📁 Project Structure

insurance-premium-predictor/
├── src/
│   ├── features/           # Feature engineering
│   │   └── feature_engineering.py
│   ├── models/            # Model implementations
│   │   └── models.py
│   ├── utils/             # Utility functions
│   │   └── metrics.py
│   └── train.py          # Training pipeline
├── data/                  # Data directory
├── submissions/          # Model predictions
└── requirements.txt      # Project dependencies

🔧 Features

Feature Engineering

• Automated categorical variable handling
• Domain-specific feature creation:
  • Income per dependent
  • Claims per year
  • Policy duration analysis

Model Implementation

• LightGBM with early stopping
• Optimized hyperparameters
• Cross-validation for robust evaluation

Quality Assurance

• Type hints throughout
• Comprehensive error handling
• Detailed logging
• Modular code structure

📊 Model Performance

The model is evaluated using 5-fold cross-validation with RMSLE as the metric:

• Mean CV RMSLE: [1.1425]
• Standard Deviation: [+/- 0.0055]

d

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

👤 Author

Your Name

• GitHub: @psukh28
• LinkedIn: surya-praanv-sukumaran

🌟 Acknowledgments

• Data source: Playground Series S4-E12
• Inspiration: Insurance premium prediction challenge
• Libraries: LightGBM, scikit-learn, pandas