Table 2 Hyperparameters Used in Training and Their Impact on Model Performance.
From: Enhancing stroke risk prediction through class balancing and data augmentation with CBDA-ResNet50
Hyperparameter | Value Tested / Used | Validation Accuracy (%) | Validation Loss |
|---|---|---|---|
Learning Rate | \(\alpha = 0.0001\) (Used) | 98.2 | 0.021 |
0.001 | 94.5 | 0.048 | |
0.00001 | 96.1 | 0.030 | |
Batch Size | 32 (Used) | 98.2 | 0.021 |
16 | 96.5 | 0.028 | |
64 | 95.8 | 0.035 | |
Class Weights | Inverse Frequency (Used) | 98.2 | 0.021 |
Equal Weights | 93.9 | 0.052 | |
Optimizer | Adam (Learning Rate: \(0.0001\), Weight Decay: \(1 \times 10^{-4}\)) | ||
Weight Decay (L2 Regularization) | \(1 \times 10^{-4}\) | ||
