Earthquake Damage Classification

Small project for predicting earthquake damage grades of buildings.

This is based on the DrivenData training competition:

https://www.drivendata.org/competitions/57/nepal-earthquake/

Model	Score
Competition top leaderboard score	0.7558
Best score achieved in this repository	0.727

The repository includes:

• Feature engineering in Python
• Random Forest baseline notebook
• XGBoost training and tuning notebooks

Project Structure

• feature_engineering.py
• 02-feature_eng.ipynb
• 03-rand_forest_classifier.ipynb
• 04-xgb_classifier.ipynb
• 07-simple_xgb-tuning.ipynb
• requirements.txt

Data Requirements

Put competition training files in data/:

• data/train_values.csv
• data/train_labels.csv

The script merges them on building_id.

Setup

1. Create and activate a virtual environment.
2. Install dependencies:

pip install -r requirements.txt

3. Start Jupyter:

jupyter notebook

Main libraries: pandas, scikit-learn, xgboost, optuna, hyperopt, mlflow, jupyter.