Table 2 Conceptual comparison between standard GOA and proposed DGOA for feature selection in DR detection.
From: An enhanced diabetic retinopathy detection approach using optimized deep learning technique
Aspect | Standard GOA | Proposed DGOA |
|---|---|---|
Parameter Control | Static control parameters | Dynamic, iteration-dependent parameter adaptation |
Exploration–Exploitation Balance | Fixed throughout optimization | Gradually transitions from exploration to exploitation |
Suitability for High-Dimensional Features | Limited due to early convergence | Enhanced via adaptive search behavior |
Feature Representation | Continuous, not task-specific | Binary encoding tailored for feature selection |
Handling Feature Redundancy | Weak pruning capability | Progressive elimination of redundant features |
Stability Across Runs | Sensitive to initialization | Improved robustness and convergence stability |
Adaptation to Medical Imaging | Generic optimization behavior | Domain-driven design for retinal fundus features |
Risk of Local Optima | Higher | Reduced through dynamic social interaction control |
