Default Risk Prediction of Enterprises Based on Convolutional Neural Network in the Age of Big Data: Analysis from the Viewpoint of Different Balance Ratios

Table 12

The comparison experiments of undersampling and oversampling on China dataset.

Method	G-mean	Accuracy	Type-II error	AUC
Undersampling
CNN	0.935	0.924	0.171	0.910
KNN	0.875	0.867	0.224	0.893
DT	0.884	0.872	0.195	0.911
SVM	0.932	0.905	0.108	0.907
LR	0.872	0.731	0.151	0.836
Oversampling
CNN	0.912	0.935	0.117	0.920
KNN	0.873	0.842	0.195	0.913
DT	0.903	0.892	0.107	0.885
SVM	0.914	0.940	0.119	0.912
LR	0.819	0.722	0.118	0.893

The bold value means that the method ranks first in some evaluation criterion. For example, CNN ranks first in terms of G-mean, accuracy, and AUC, which indicates that its comprehensive ranking also ranks first.