Journal of Big Data

Table 7 Average results by learner and evaluation metric for a subset of papers, experimental/novel methods vs. baseline methods

From: The use of generative adversarial networks to alleviate class imbalance in tabular data: a survey

	AUC	Balanced Accuracy	F1 Score	Precision	Ranking	Recall	novel > baseline	baseline > novel
CNN							4	0
Novel		*0.6767*	*0.9947*	*0.9936*		*0.9958*
Baseline		0.6187	0.9783	0.9792		0.9805
DT							2	0
Novel			*0.9552*		*3.3370*
Baseline			0.9482		4.0549
GBC							1	0
Novel					*2.4850*
Baseline					4.2097
KNN							1	1
Novel			0.2600		*2.8140*
Baseline			*0.2736*		4.1378
LR							1	0
Novel					*3.6300*
Baseline					3.9697
MLP							3	1
Novel		*0.9937*	*0.9935*	*0.9890*		0.9637
Baseline		0.9723	0.9703	0.9725		*0.9814*
NB							0	1
Novel			0.1800
Baseline			*0.4279*
RF							2	0
Novel			*0.9543*		*2.5857*
Baseline			0.9485		4.0488
SAE							4	0
Novel		*0.9951*	*0.9947*	*0.9936*		*0.9958*
Baseline		0.9805	0.9783	0.9792		0.9805
SVM							1	2
Novel	0.7128		0.4275		*2.4633*
Baseline	*0.9412*		*0.5231*		4.2556
novel > baseline	0	3	5	3	6	2	19
baseline > novel	1	0	3	0	0	1		5

Experimental methods have a "better" evaluation metric in 19 of 24 aggregation instances

Back to article page