Skip to content

Best Model Selection with Optuna

Results

Optuna Results Comparison

XGBoost wins again with AUC 0.968 (improved from V1's 0.966)

Other results: - CatBoost: AUC 0.967 - LightGBM: AUC 0.967

Winning XGBoost Parameters

XGBoost Optuna Parameters

Class Imbalance Impact

The winning model shows a high number of false negatives, likely due to the 7.3:1 class imbalance in the dataset. While achieving strong AUC performance, the model tends to under-predict positive cases (subscription = 1). This issue will be addressed in V4 through class imbalance handling techniques.