Table 6.

Performance Metrics of model ResNet convolutional neural network (CNN).

Evaluation Metrics of ResNet-14 CNN
	Multi Classes
Approaches	Evaluation	Healthy Tear	Partial Tear	Full Torn	Average
Without class balancing and data augmentation (5-fold cross-validation)	Precision	0.85	0.57	0.57	0.663
	Sensitivity	0.96	0.39	0.22	0.523
	F1-Score	0.90	0.47	0.31	0.563
	Specificity	0.78	0.86	0.95	0.863
	Accuracy	0.81
	AUC	0.87	0.81	0.91	0.863
Without class balancing with data augmentation (5-fold cross validation)	Precision	0.83	0.47	0.47	0.590
	Sensitivity	0.94	0.29	0.22	0.483
	F1-Score	0.88	0.36	0.30	0.513
	Specificity	0.70	0.78	0.96	0.813
	Accuracy	0.77
	AUC	0.83	0.76	0.91	0.833
Hybrid class balancing without data augmentation (Random Splitting)	Precision	0.87	0.81	0.96	0.880
	Sensitivity	0.85	0.79	0.99	0.877
	F1-score	0.86	0.80	0.98	0.880
	Specificity	0.90	0.92	0,99	0.910
	Accuracy	0.88
	AUC	0.96	0.95	0.99	0.967
Hybrid class balancing with data augmentation (random splitting)	Precision	0.89	0.84	0.94	0.890
	Sensitivity	0.86	0.81	0.99	0.887
	F1- score	0.88	0.83	0.97	0.893
	Specificity	0.91	0.92	0.99	0.940
	Accuracy	0.90
	AUC	0.97	0.96	0.99	0.973
Hybrid class balancing with data augmentation (3-fold cross validation)	Precision	0.90	0.83	0.94	0.890
	Sensitivity	0.87	0.80	0.99	0.887
	F1- score	0.88	0.82	0.97	0.890
	Specificity	0.91	0.92	0.99	0.940
	Accuracy	0.90
	AUC	0.97	0.94	0.99	0.967
Hybrid class balancing with data augmentation (5-fold cross validation)	Precision	0.92	0.87	0.96	0.917
	Sensitivity	0.89	0.87	0.99	0.917
	F1-score	0.90	0.87	0.98	0.917
	Specificity	0.93	0.92	0.99	0.947
	Accuracy	0.92
	AUC	0.98	0.97	0.99	0.980