Table 15 Comprehensive before/after classification performance comparison across GAN architectures. Significant values are in bold.
GAN architecture | Avg precision before | Avg Precision After | Precision improvement (%) | Avg recall before | Avg recall after | Recall improvement (%) | Avg F1-score before | Avg F1-score after | F1-score improvement (%) |
|---|---|---|---|---|---|---|---|---|---|
Dual-Gland GAN (Ours) | 0.842 | 0.916 | + 8.8 | 0.854 | 0.929 | + 8.8 | 0.847 | 0.922 | + 8.9 |
Traditional GAN | 0.842 | 0.867 | + 3.0 | 0.854 | 0.871 | + 2.0 | 0.847 | 0.869 | + 2.6 |
DCGAN | 0.842 | 0.879 | + 4.4 | 0.854 | 0.886 | + 3.7 | 0.847 | 0.882 | + 4.1 |
WGAN | 0.842 | 0.891 | + 5.8 | 0.854 | 0.897 | + 5.0 | 0.847 | 0.894 | + 5.5 |
Conditional GAN | 0.842 | 0.883 | + 4.9 | 0.854 | 0.889 | + 4.1 | 0.847 | 0.886 | + 4.6 |
CycleGAN | 0.842 | 0.888 | + 5.5 | 0.854 | 0.894 | + 4.7 | 0.847 | 0.891 | + 5.2 |
GSIP-GAN | 0.842 | 0.902 | + 7.1 | 0.854 | 0.908 | + 6.3 | 0.847 | 0.905 | + 6.8 |
ECP-IGANN | 0.842 | 0.908 | + 7.8 | 0.854 | 0.915 | + 7.1 | 0.847 | 0.911 | + 7.6 |
MCI-GAN | 0.842 | 0.895 | + 6.3 | 0.854 | 0.901 | + 5.5 | 0.847 | 0.898 | + 6.0 |
