Table 10 Comparative feasibility of transformer-based models vs. proposed EDHL.
Model Type | Accuracy | Memory usage (MB) | Runtime (s) |
|---|---|---|---|
CNN (Baseline) | 92% | 490 | 55 |
ResNet | 93% | 510 | 56 |
Hybrid (ResNet–LightGBM) | 94% | 480 | 58 |
Transformer-Based (ViT, Swin) | 97% | 1200 | 95 |
Proposed EDHL (Inception–GB) | 98% | 455 | 49.5 |
