Leveraging augmentation techniques for tasks with unbalancedness within the financial domain: a two-level ensemble approach

EPJ Data Science

Table 17 Classification results for the credit card frauds dataset

Evaluation metric	Naive Bayes	SGD	KNN	MLP	Random forest	SVM
Without augmentation
F1 score macro	0.8218	0.3373	0.3351	0.3395	0.8781	0.3351
F1 score micro	0.8277	0.5052	0.5042	0.5062	0.8802	0.5042
Precision minority class	0.9942	0	0	0	1	0
Recall minority class	0.6566	0.002	0	0.004	0.7587	0
Baseline: min-max approach
F1 score macro	0.7282	0.3351	0.3351	0.3351	0.8615	0.3351
F1 score micro	0.7471	0.5042	0.5042	0.5042	0.8647	0.5042
Precision minority class	1	0	0	0	1	0
Recall minority class	0.4902	0	0	0	0.7266	0
Baseline: mean-std approach
F1 score macro	0.5465	0.3351	0.3351	0.3351	0.3788	0.3351
F1 score micro	0.615	0.5042	0.5042	0.5042	0.5244	0.5042
Precision minority class	0.9726	0	0	0	1	0
Recall minority class	0.2297	0	0	0	0.041	0
SMOTE approach
F1 score macro	0.8847	0.7158	0.9849	0.9485	0.976	0.5409
F1 score micro	0.8862	0.7605	0.9849	0.9486	0.976	0.5528
Precision minority class	0.9923	0.8797	0.9819	0.9867	0.9936	0.5087
Recall minority class	0.7769	0.7316	0.9878	0.909	0.958	0.5015
SMOTE-clustering approach
F1 score macro	0.8798	0.8812	0.9879	0.9395	0.9898	0.549
F1 score micro	0.8814	0.8833	0.9879	0.9397	0.9898	0.5584
Precision minority class	0.9822	0.9887	0.9901	0.9871	0.9963	0.5127
Recall minority class	0.7759	0.775	0.9857	0.8908	0.9832	0.5454