Richter's Predictor: Modeling Earthquake Damage [Link]

We're trying to predict the ordinal variable damage_grade, which represents a level of damage to the building that was hit by the earthquake. This is me: Shoumik

Solution [Notebook]

This notebook contains the current best scoring solution. A score of 0.7539 on drivendata.org

Feature Extraction [Notebook]

This answers the following questions:

1. Geo Dimensionality Reduction: How to Use Neural Network Dimensionality Reduction to get dense embeddings out of Geo Level IDs
2. Geo 3 Rollup: How to obtain dense embeddings from Geo Level IDs by treating geo levels as a supervised training problem of predicting geo level IDs 1 and 2 from Geo Level 3 ID as input.

XGBoost [Notebook]

Using Feature Extraction and Hyperparameter Optimization using Optuna, A well performing XGBoost Model is obtained.
A Bagging Model is attempted.

CatBoost [Notebook]

Using Feature Extraction and Hyperparameter Optimization using Optuna, A well performing CatBoost Model is obtained.

LightGBM [Notebook]

Using Feature Extraction and Hyperparameter Optimization using Optuna, A well performing LightGBM Model is obtained.