Table 4 Summary of key experiments.
From: An enhanced diabetic retinopathy detection approach using optimized deep learning technique
Experiment | Objective | Key metrics |
|---|---|---|
Baseline Comparison | Compare DGOA-ensemble with existing traditional optimization and classification methods | Accuracy, F1-score, AUC-ROC, inference time |
Ablation Study | Assess DGOA’s impact on feature selection | Accuracy change, feature reduction, training time |
Dataset Split Testing | Evaluate generalization ability | Cross-validation results, standard deviation of accuracy |
Noisy Data & Augmentation | Test robustness against real-world distortions | Accuracy drop, robustness factor |
Computational Efficiency | Measure scalability and speed | Training time, inference speed, memory usage |
Class Imbalance Handling | Evaluate performance on minority classes | AUC-ROC, Precision-Recall, F1-score |
Statistical Significance | Confirm reliability of improvements | p-value, confidence intervals |
Explainability Analysis | Understand feature importance | SHAP/LIME visualizations, feature impact scores |
Comparative Analysis | Compare the suggested model with state-of-the art diabetic retinopathy detection approaches | Accuracy, F1-score, AUC-ROC |
