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 13

The comparison experiments of feature selection methods on China dataset.

Method	G-mean	Accuracy	Type-II error	AUC
FDAF-score
CNN	0.904	0.927	0.124	0.921
KNN	0.868	0.845	0.177	0.891
DT	0.895	0.900	0.121	0.899
SVM	0.900	0.914	0.116	0.904
LR	0.857	0.876	0.145	0.884
Correlation coefficient
CNN	0.920	0.912	0.102	0.931
KNN	0.894	0.882	0.178	0.921
DT	0.903	0.904	0.131	0.864
SVM	0.923	0.900	0.115	0.922
LR	0.891	0.895	0.097	0.924
Particle swarm optimization algorithm
CNN	0.929	0.935	0.101	0.935
KNN	0.867	0.852	0.203	0.908
DT	0.907	0.912	0.147	0.879
SVM	0.924	0.934	0.119	0.921
LR	0.921	0.927	0.093	0.893
Artificial bee colony algorithm
CNN	0.941	0.925	0.067	0.935
KNN	0.888	0.929	0.178	0.894
DT	0.915	0.878	0.187	0.917
SVM	0.921	0.930	0.075	0.942
LR	0.845	0.884	0.238	0.876

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.