Table 6 Ablation study of the proposed model.
From: A hybrid XAI-driven deep learning framework for robust GI tract disease diagnosis
Ablation experiment | Description | Accuracy (%) | Precision | Recall | F1-score | MCC | Train time (s) | Infer time (s) |
|---|---|---|---|---|---|---|---|---|
Hybrid model (without stacking) | All components are included but results are obtained without stacking | 92.93 | 0.9327 | 0.9293 | 0.9296 | 0.9195 | 3676.63 | 0.1888 |
Hybrid model (with stacking) | All components are included but results are obtained by stacking | 92.59 | 0.9343 | 0.9351 | 0.9322 | N/A | 3676.63 | 0.1889 |
EfficientNet | Remove Swin transformer and ResNet | 87.75 | 0.8989 | 0.8775 | 0.8759 | 0.8637 | 1937.42 | 0.0140 |
ResNet | Remove Swin transformer and EfficientNet | 91.16 | 0.9172 | 0.9116 | 0.9116 | 0.8998 | 1936.14 | 0.0066 |
Swin transformer | Remove DCNN and meta-classifier | 11.87 | 0.0141 | 0.1187 | 0.0252 | 0.0000 | 2109.64 | 0.0110 |
