Richter’s Predictor: Modeling Earthquake Damages

The report presents an algorithmic approach to tackle the Richter’s Predictor: Modeling Earthquake Damage problem, achieving an impressive f1-score of 0.7525.The algorithm incorporates a range of data preprocessing techniques, including feature selection based on mutual information, log transformation to normalize numerical features, geological feature encoding using a neural network embedding,and Principal Component Analysis to reduce the dataset’s dimensionality. LightGBM is selected as the primary training model following a rigorous evaluation of its performance compared to other models such as Neural Network, XGBoost,and Catboost. Lastly, Ensemble learning is utilized to combine the predictions of multiple models, thereby enhancing the algorithm’s generalization ability. You can find more details here.