Table 3:

5-fold cross validation performance of the proposed Multi-task model, the CNN model, and the Grassmann model, on the AREDS and AREDS2 datasets.

	AREDS		AREDS2
	Multi-task	CNN	Multi-task	CNN	Grassmann
Precision	0.591	0.592	0.545	0.533	0.514
Recall	0.614	0.612	0.472	0.437	0.345
F1-score	0.597	0.593	0.492	0.449	0.398
Kappa	0.487	0.484	0.330	0.297	0.216
Accuracy	0.614	0.612	0.472	0.437	0.346