churn-prediction

Testing neural and tree ML algorithms on a customer Churn dataset.

The notebook includes data preparation, data separation, training and tests with the following algorithms:

• MLP
• Random Forest
• KAN
• XGBoost
• CatBoost
• STab
• TabKANet
• TabPFN

We used Optuna + cross validation for most of the algorithms, reaching a 0.586 KS statistic and 0.86 AUC ROC using TabPFN:

Contributors: Fábio Papais, Jaubert Gualberto and Silvânio Assunção