Table 10. Statistical summary of 12 DNN models trained to distinguish Abs from a specific family lineage using the Robust Convolutional Autoencoder one-class classification method.

DNN model	Training and Validation Image Set Sizes				Testing Image Set & Results
	Ntraininga	Mvalidationb		AUROC Trainc	Ntestd		AUROC Teste	normal Ab Id
	normal	normal	anomalous		normal	anomalous
1	765	135	5700	0.84	300	1500	0.95	ADI-15841
2	765	135	5700	0.72	300	1500	0.65	ADI-15916
3	765	135	5700	0.74	300	1500	1.00	ADI-15925
4	765	135	5700	0.84	300	1500	0.79	ADI-15785
5	765	135	5700	0.99	300	300	0.45	ADI-15935
6	765	135	5700	0.89	300	300	0.81	ADI-15772
7	765	135	5700	0.94	300	300	0.14	ADI-15780
8	765	135	5700	0.61	300	300	1.00	ADI-15784
9	442	78	3800	1.00	200	1000	0.99	ADI-15843
10	510	90	3800	1.00	120	1000	0.89	ADI-15912
11	442	78	3800	1.00	200	1000	1.00	ADI-15978
12	442	78	3800	0.84	200	1000	0.98	ADI-15861
Average (SD)				0.87 (0.13)			0.80 (0.27)

^a N_training; number of fingerprints from normal Abs selected for training of the model.

^b M_validation; number of fingerprints from normal and anomalous Ab classes selected for validation.

^c AUROC Train is computed on the training set using the Python Scikit-learn library for machine learning and statistical modeling [25].

^d N_test; number of fingerprints from normal and anomalous Ab classes in the testing sets.

^e AUROC Test is computed on the testing set using the Python Scikit-learn library for machine learning and statistical modeling [25].