A framework to improve churn prediction performance in retail banking

Financial Innovation

Table 5 Average of the 100-fold cross validation predictive performance for the 18 assessed configurations

Configuration	FE	IDT-Over	IDT-Under	Classification model	PR-AUC	Accuracy	Recall	Specificity
C1	Yes	ADASYN	NEARMISS	XGBoost	0.9518	0.8860	0.8967	0.8884
C2	Yes	ADASYN	NEARMISS	Elastic Net	0.9157	0.8632	0.8654	0.8592
C3	Yes	ADASYN	RU	XGBoost	0.8857	0.8194	0.8401	0.8184
C4	Yes	None	NEARMISS	Elastic Net	0.8843	0.8208	0.8509	0.8225
C5	No	ADASYN	NEARMISS	Elastic Net	0.8614	0.7769	0.7802	0.7755
C6	Yes	None	RU	Elastic Net	0.8610	0.8045	0.8199	0.8041
C7	Yes	ADASYN	RU	Elastic Net	0.8564	0.7845	0.8300	0.7890
C8	No	None	NEARMISS	Elastic Net	0.8397	0.7435	0.7623	0.7429
C9	No	ADASYN	RU	Elastic Net	0.8110	0.6843	0.7655	0.6856
C10	No	None	RU	Elastic Net	0.8103	0.7128	0.7531	0.7141
C11	Yes	None	RU	XGBoost	0.8025	0.7444	0.7544	0.7468
C12	Yes	None	NEARMISS	XGBoost	0.7987	0.7399	0.7549	0.7445
C13	No	ADASYN	NEARMISS	XGBoost	0.7055	0.6337	0.6435	0.6320
C14	No	ADASYN	RU	XGBoost	0.6543	0.5671	0.6095	0.5630
C15	No	None	NEARMISS	XGBoost	0.5934	0.5263	0.5495	0.5283
C16	No	None	RU	XGBoost	0.5870	0.5295	0.5337	0.5301
C17	No	None	None	XGBoost	0.5000	0.9989	0.0000	1.0000
C18	No	None	None	Elastic Net	0.5000	0.9988	0.0000	1.0000