Table 5. Results of the ablation study.

Methods	Model			Recall	F1-Score	AUC	G-mean
	GBDT	NCL	CTGAN
TCN(O)	–	–	–	0.7407	0.7717	0.8702	0.8605
GBDT(O)	✓	–	–	0.7469	0.8013	0.8733	0.8641
CTGAN(O)	–	–	✓	0.7839	0.8015	0.8918	0.8852
OS(N)	–	✓	✓	0.8024	0.8049	0.9010	0.8956
CTCN	✓	✓	✓	0.8299	0.8187	0.9147	0.9108

Notes.

The values in bold indicate the best results.

TCN(O) denotes that the original data is detected using TCN; GBDT(O) denotes that the original data is feature-selected and then detected using TCN; CTGAN(O) denotes that CTGAN balances the original dataset and then detects it using TCN; OS(N) denotes that the original dataset is balanced using improved CTGAN and then TCN is used for detection; CTGAN indicates that the full module was used.