From: Measuring the model risk-adjusted performance of machine learning algorithms in credit default prediction
Logit
Lasso
CART
Random forest
XGBoost
MLP
AUC-ROC
80%
80.2%
80.4%
84.2%
85.3%
81.7%
95% interval
79.3%
80.7%
79.6%
80.8%
79.7%
81.2%
83.7%
84.8%
85.8%
81%
82.3%
AUC-ROC (difference with logit)
–
0.2%
0.4%
4.2%
5.3%
1.7%