Table 5 Performance vs. activation function (3-layer, \(50\%\) dropout).
From: Embedding-driven dual-branch approach for accurate breast tumor cellularity classification
Activation | Accuracy (%) | Sensitivity (%) | F1 (%) | Average (%) |
|---|---|---|---|---|
ReLU | 96.43 | 96.43 | 96.43 | 96.98 |
Sigmoid | 92.86 | 92.86 | 92.86 | 93.21 |
Tanh | 95.00 | 95.00 | 95.00 | 95.42 |
LeakyReLU (0.01) | 97.86 | 97.86 | 97.86 | 98.15 |
Swish | 97.14 | 97.14 | 97.14 | 97.52 |
