Table 2 Classification performance based on vectorization method.

Method	ER+		ER−		PR+		PR−		HER2+		HER2−
Method	F1	AUC	F1	AUC	F1	AUC	F1	AUC	F1	AUC	F1	AUC
RWOV-NN	0.95	0.95	0.82	0.96	0.93	0.96	0.86	0.95	0.68	0.89	0.91	0.86
RWOV-SVM	0.90	0.88	0.70	0.88	0.76	0.72	0.58	0.71	0.53	0.77	0.84	0.77
SVM(1,2)	0.92	0.94	0.69	0.94	0.91	0.95	0.81	0.95	0.31	0.67	0.87	0.68
SVM(2,2)	0.93	0.94	0.72	0.94	0.91	0.96	0.82	0.96	0.37	0.73	0.90	0.74
SVM(1,3)	0.94	0.94	0.73	0.94	0.90	0.95	0.80	0.95	0.32	0.69	0.89	0.70
SVM(2,3)	0.94	0.94	0.75	0.94	0.91	0.95	0.82	0.95	0.40	0.72	0.91	0.74
SVM(3,3)	0.93	0.92	0.71	0.92	0.90	0.92	0.79	0.92	0.35	0.71	0.91	0.72
SVM-W2V	0.71	0.64	0.40	0.64	0.67	0.69	0.57	0.69	0.20	0.45	0.60	0.43
NN(1,2)	0.93	0.95	0.71	0.94	0.90	0.95	0.82	0.95	0.35	0.81	0.91	0.79
NN(2,2)	0.93	0.93	0.72	0.94	0.92	0.96	0.81	0.95	0.39	0.80	0.91	0.80
NN(1,3)	0.94	0.94	0.72	0.95	0.91	0.95	0.81	0.95	0.30	0.79	0.91	0.79
NN(2,3)	0.93	0.92	0.70	0.93	0.92	0.94	0.80	0.95	0.33	0.78	0.90	0.78
NN(3,3)	0.93	0.92	0.69	0.91	0.89	0.92	0.79	0.92	0.20	0.76	0.90	0.76
NN-W2V	0.85	0.76	0.40	0.75	0.81	0.75	0.63	0.78	0.07	0.43	0.87	0.41

RWOV had the most consistent performance across classification tasks. The best performing method for each metric on each task is shown in bold in the table. In only one case did RWOV-NN not have the best performance (PR− AUC), however it was very close to the top performer.

Quick links

Search