Building Damage Prediction
Description
A DrivenData.org machine learning competition to predict damage to buildings from a 2015 earthquake from 1 (low damage) to 3 (total destruction)
Goal
Acheive highest possible micro-averaged F1 score on unseen test data.
Outcome
Best submission acheived a micro-averaged F1 score of .7457
Notable Features
Best submission ranks in the top 5% of submissions to date.
Technology Used
- Python
- scikit-learn
- XGBoost
- SMOTE
- Tomek Links Undersampling
- Target encoding
- Iterative imputation