Table 9.

Evaluation metrics of our proposed model using Vahadane normalization

Folds	Confusion matrices					Performance evaluation
	Predict $\to$ Actual $\downarrow$	Ben.	Ins.	Inv.	Nor.	Prec.	Rec.	F1	Test	Accuracy (%)	Kappa
Fold 1	Benign	42	4	2	2	1.00	0.84	0.91	50	96.66	0.948
	In situ	0	49	1	0	0.91	0.98	0.94	50
	Invasive	0	0	226	4	0.98	0.98	0.98	230
	Normal	0	1	1	118	0.95	0.98	0.97	120
Fold 2	Benign	45	2	0	3	0.98	0.90	0.94	50	96.44	0.945
	In situ	1	48	0	1	0.91	0.96	0.93	50
	Invasive	0	2	222	6	1.00	0.97	0.98	230
	Normal	0	1	0	119	0.92	0.99	0.96	120
Fold 3	Benign	46	1	1	2	0.98	0.92	0.95	50	95.77	0.934
	In situ	1	48	1	0	0.96	0.96	0.96	50
	Invasive	0	0	227	3	0.95	0.99	0.97	230
	Normal	0	1	9	110	0.96	0.92	0.94	120
Fold 4	Benign	47	1	0	2	0.98	0.94	0.96	50	97.77	0.965
	In situ	1	48	1	0	0.96	0.96	0.96	50
	Invasive	0	0	227	3	0.99	0.99	0.99	230
	Normal	0	1	1	118	0.96	0.98	0.97	120
Fold 5	Benign	48	1	1	0	0.89	0.96	0.92	50	96.22	0.942
	In situ	2	48	0	0	0.94	0.96	0.95	50
	Invasive	0	1	223	6	0.99	0.97	0.98	230
	Normal	4	1	1	114	0.95	0.95	0.95	120
Final	Benign	46	1	1	2	0.98	0.92	0.95	50	97.33	0.958
	In situ	1	47	2	0	0.96	0.94	0.95	50
	Invasive	0	0	226	4	0.99	0.98	0.98	230
	Normal	0	1	0	119	0.95	0.99	0.97	120