Table 9 Generalization performance across different dataset splits and cross-validation settings.
From: An enhanced diabetic retinopathy detection approach using optimized deep learning technique
Dataset | Train-Test Split | Accuracy (%) | Std. Dev. | Variance | 5-Fold CV Accuracy (%) | 10-Fold CV Accuracy (%) |
|---|---|---|---|---|---|---|
DIARETDB1 | 80 − 20 | 88.5 | 2.1 | 0.044 | 86.8 | 87.4 |
70 − 30 | 86.2 | 3.5 | 0.123 | 85.5 | 86.1 | |
60 − 40 | 84.8 | 4.2 | 0.176 | 84.1 | 84.6 | |
MESSIDOR | 80 − 20 | 91.2 | 1.8 | 0.032 | 90.5 | 90.9 |
70 − 30 | 89.7 | 2.7 | 0.073 | 89.1 | 89.5 | |
60 − 40 | 88.5 | 3.1 | 0.096 | 88.0 | 88.3 | |
DRIVE | 80 − 20 | 92.8 | 1.4 | 0.019 | 91.9 | 92.3 |
70 − 30 | 91.1 | 2.2 | 0.048 | 90.7 | 91.0 | |
60 − 40 | 89.8 | 3.0 | 0.090 | 89.2 | 89.5 | |
APTOS 2019 | 80 − 20 | 87.4 | 2.5 | 0.063 | 86.9 | 87.1 |
70 − 30 | 85.9 | 3.2 | 0.102 | 85.5 | 85.7 | |
60 − 40 | 84.3 | 4.1 | 0.168 | 83.8 | 84.0 |
