amex-default-prediction

Our team's solution to a Machine Learning Comp hosted by American Express.

Preprocessing and sampling

• subsample_data.py: Randomly selects CIDS and all of their monthly statements from the provided data
• shrinkData.py: Averages the monthly statements of each CID and returns a specified amount of that data
• preprocess.py: Performs feature engineering as described in report

Metric

• amex_metric.py: Contains the calculations for the metric used for scoring in the competition

Model Training

• modelTrain.py: Trains the three models specified within report (XGBoost, Catboost, LightGBM), providing CV score and saves these models

Prediction

• predictLgb.py: Loads provided lgb model and makes prediction on the provided data
• predictXgb.py: Loads provided Xgb model and makes prediction on the provided data
• predictCb.py: Loads provided Cb model and makes prediction on the provided data

Hyperparameter Tuning

• lgbmTuning.py: Tunes the hyperparameters of a LGBM model
• xgbTuning.py: Tunes the hyperparameters of a LGBM model
• cbTuning.py Tunes the hyperparameters of a LGBM model

Processed Dataset

• Processed Data

Future Works

• permutation_importance.py: Permutation importance to improve on dataset processing