Team: ES - Igor Vdovin, Timur Enikeev, Nikita Usachev, Sergey Sharkov (me)
Tools: LightGBM, CatBoost, XGBoost, AutoML, Scikit-Learn, Pandas, Optuna, Shap
Concepts: Ensembling, Cross-validation, Classifier calibration, Extracting business features, Feature importance analysis, Time Series Analysis
Topic: Customer churn prediction. [Case Link]
This repository contains the top-1 solution for Siberian Alpha Hack 2023. Our team implemented ensemble of gradient boosting models, tuned with optuna. Also we implemented decision tree on self generated "indicator-like" features as interpretable model for business.
We performed calibration of our classifier based on synthetic example by applying different cost for FP/FN types errors (more info can be found in our presentation DefendencePresentation.pdf)
In terminal execute the following command with specified arguments:
python inference.py --path_to_models_dir "models" --path_to_test_parquet "data/test.parquet"