Table 3.
Summary of K-fold cross-validation over the experimental models (ResNet50, InceptionV3, and VGG16) shows accuracy mean and standard deviation recorded among 10-folds, 5-folds, and 3-folds cross-validation for every trained model using SGD and Adam optimizers.
| SA + ResNet50 | SA + VGG16 | SA + InceptionV3 | ||||
|---|---|---|---|---|---|---|
| Mean | STD | Mean | STD | Mean | STD | |
| 10-Folds | ||||||
| Adam | 99.8% | +/- 0.38% | 47.93% | +/- 26.46% | 27.24% | +/- 10.59% |
| SGD | 96.55% | +/- 1.26% | 59.54% | +/- 37.00% | 96.44% | +/- 1.20% |
| 5-Folds | ||||||
| Adam | 88.97% | +/- 1.17% | 71.72% | +/- 9.35% | 28.89% | +/-8.60% |
| SGD | 96.55% | +/- 1.03% | 97.01% | +/- 1.56% | 96.09% | +/- 0.56% |
| 3-Folds | ||||||
| Adam | 87.36% | +/- 1.63% | 73.56% | +/- 5.71% | 61.30% | +/- 21.53% |
| SGD | 95.02% | +/- 1.05% | 73.56% | +/- 7.68% | 95.79% | +/- 1.43% |