CarSite-II: an integrated classification algorithm for identifying carbonylated sites based on K-means similarity-based undersampling and synthetic minority oversampling techniques

Table 1 Comparison of different resampling methods on our independent test data

Resample method	Sn (%)	Sp (%)	Acc (%)	Mcc	AUC	G-mean
K
Without resampling	5.13	95.16	93.77	0.0017	0.4959	0.2209
SMOTE	41.88	98.43	97.55	0.3395	0.8868	0.6420
KSU undersampling	70.94	86.54	86.30	0.2025	0.8096	0.7835
CarSite-II	89.74	98.35	98.21	0.6358	0.9603	0.9395
P
Without resampling	0	100	99.70	NaN	0.6116	0
SMOTE	50.00	97.61	97.47	0.1658	0.8512	0.6986
KSU undersampling	31.25	99.64	99.44	0.2524	0.8810	0.5580
CarSite-II	81.25	97.97	97.92	0.2910	0.8768	0.8922
R
Without resampling	3.70	96.65	95.81	0.0018	0.6210	0.1892
SMOTE	27.78	97.96	97.33	0.1627	0.8695	0.5216
KSU undersampling	46.30	88.18	87.81	0.0996	0.7631	0.6389
CarSite-II	79.63	98.12	97.96	0.4629	0.9236	0.8839
T
Without resampling	8.33	98.43	98.10	0.0327	0.7539	0.2864
SMOTE	12.50	98.36	98.04	0.0510	0.8250	0.3506
KSU undersampling	45.83	92.29	92.11	0.0857	0.8120	0.6504
CarSite-II	66.67	99.06	98.94	0.3685	0.8602	0.8127

ISSN: 1471-2105