Table 12 Impact of SMOTE and hyperparameter tuning on mild DR detection performance on retinal fundus images dataset.
From: An enhanced diabetic retinopathy detection approach using optimized deep learning technique
Dataset version | AUC-ROC | F1-Score (Mild DR) | Precision (Mild DR) | Recall (Mild DR) | Interpretation |
|---|---|---|---|---|---|
Original (Imbalanced) | 0.78 | 0.52 | 0.60 | 0.45 | The model struggles with mild DR cases, leading to lower recall due to class imbalance. It tends to favor the majority class. |
SMOTE-Balanced | 0.84 | 0.68 | 0.65 | 0.72 | Oversampling improves recall for mild DR cases, helping the model detect more positive instances correctly while maintaining precision. |
SMOTE + Tuning | 0.86 | 0.74 | 0.70 | 0.78 | Additional hyperparameter tuning enhances performance, achieving a better balance between precision and recall for mild DR. |
