Table 9 Numerical results of classification experiments integrating discriminative feature selection and random forest.
From: An efficient method for identifying surface damage in hydraulic concrete buildings
Condition | With random forest | With feature selection & random forest | ||||||
|---|---|---|---|---|---|---|---|---|
Models | ResNet | MobileNet | EfficientNet | RegNetY | ResNet | MobileNet | EfficientNet | RegNetY |
Trainable parameters | 4722692 | 1484164 | 1134516 | 1593796 | 4722692 | 1484164 | 1134516 | 1593796 |
Accuracy | 0.89 | 0.8855 | 0.911 | 0.9102 | 0.892 | 0.8932 | 0.9142 | 0.917 |
Precision | 0.8889 | 0.8855 | 0.9106 | 0.91 | 0.8911 | 0.8945 | 0.9139 | 0.9167 |
Recall | 0.89 | 0.8855 | 0.911 | 0.9102 | 0.892 | 0.8932 | 0.9142 | 0.917 |
F1-Score | 0.889 | 0.8834 | 0.91 | 0.91 | 0.8913 | 0.8921 | 0.9133 | 0.9167 |
Original dimension | 512 | 960 | 1280 | 768 | 512 | 960 | 1280 | 768 |
Final dimension | 512 | 960 | 1280 | 768 | 280 | 240 | 140 | 280 |
Training Time(s) | 4.79 | 4.67 | 12.53 | 5.61 | 8.83 | 6.74 | 10.14 | 12.06 |
Infer time (s) | 1.62 | 1.71 | 1.8 | 1.27 | 0.05 | 0.1 | 0.1 | 0.0324 |
