Table 5 Summary of key parameters used in model training and evaluation.
From: Comparative analysis of automated foul detection in football using deep learning architectures
Parameter | Value / Description |
|---|---|
Input Image Size | 224 × 224 pixels |
Batch Size | 32 |
Image Normalization | Rescaled to [0, 1] |
Optimizer | Adam |
Learning Rate | 0.0001 |
Loss Function | Binary Crossentropy |
Primary Metric | Accuracy |
Secondary Metrics | Precision, Recall, F1 Score, AUC |
Maximum Epochs | 300 |
Early Stopping Patience | 10 epochs |
Model Checkpointing | Save model with lowest validation loss |
Explainability Technique | GradCAM++ |
GradCAM + + Samples per Epoch | 5 per training, validation, and test sets |
Test set balance | 350 Foul, 350 Not Foul (where feasible) |
Train/Validation/Test Split | 70% / 20% / 10% |
Data Pipeline Framework | TensorFlow tf.data |
