Table 4.
The performance of all Ml models and DNN for detection of DR using a combination of VGG-16 and retinal lesion features.
| Classifier | Accuracy % | Precision | Recall | F1-score | AUC |
|---|---|---|---|---|---|
| Xgboost | 99.38 | 1 | 0.99 | 0.99 | 0.994 |
| KNN | 47.53 | 0.45 | 0.25 | 0.32 | 0.475 |
| LR | 51.85 | 0.52 | 0.57 | 0.54 | 0.519 |
| SVM | 43.21 | 0.45 | 0.62 | 0.52 | 0.432 |
| MLP | 53.09 | 0.52 | 0.93 | 0.66 | 0.531 |
| DT | 88.27 | 1.00 | 0.77 | 0.87 | 0.883 |
| RF | 68.52 | 0.71 | 0.63 | 0.67 | 0.685 |
| DNN (1 layer) | 48 | 0.48 | 0.58 | 0.53 | 0.481 |
| DNN (2 layers) | 50 | 0.5 | 0.38 | 0.43 | 0.5 |
| DNN (3 layers) | 52 | 0.52 | 0.56 | 0.54 | 0.519 |
The bold values highlight the best accuracy, precision, recall, F1-score, and AUC of ML and DNN learning models for detection of DR using a combination of VGG-16 and retinal lesion features.