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

Journal of Big Data

Table 2 Average results of experiments related to financial transactions, aggregated by paper, novel vs. baseline, and sampling method

Paper - Novel vs Baseline - Sampling Method	AUC	Balanced Accuracy	F1 Score	Ranking	Recall	Specificity
Conditional Wasserstein GAN-based oversampling of tabular data for imbalanced learning
Novel
CWGAN				3.15
Baseline
ADAYSN				5.37
B-SMOTE				4.03
None				*2.32*
Random				3.48
SMOTE				3.90
SMOTE-ENC				5.73
SMOTENC				5.17
Generative adversarial fusion network for class imbalance credit scoring
Novel
IGAFN	*71.08%*	*83.60%*	*57.98%*
Baseline
CFN	65.64%	76.05%	47.94%
GAN	70.07%	79.39%	54.43%
No treatment	63.55%	79.35%	42.12%
SMOTE	65.38%	70.98%	43.27%
Using generative adversarial networks for improving classification effectiveness in credit card fraud detection
Novel
GAN					*71.94%*	99.99%
Baseline
SMOTE					70.60%	*100.00%*

Average results are displayed by evaluation metric, and top performers are bolded and italicized