Focal Loss Parameter Performance for Peritoneal Metastasis Prediction Model

Chao Zhang published a 9.8 KB table on May 14, 2026, detailing hyperparameter tuning results for a computer vision model. The data shows the training and validation ROC-AUC scores for an Inception-ResNetV2-SE model under different Focal Loss settings. It identifies the optimal parameter combination of α=0.75 and γ=2, which achieved a validation ROC-AUC of 0.9531.

Use Cases

• Comparing the impact of different Focal Loss hyperparameters on model validation performance.
• Identifying optimal α and γ values for training models on imbalanced medical image datasets.
• Benchmarking the Inception-ResNetV2-SE architecture's performance for a specific medical prediction task.
• Analyzing the relationship between training epoch and validation ROC-AUC to prevent overfitting.

Strengths

• The dataset clearly identifies the best-performing parameter combination (α=0.75, γ=2) with a concrete validation ROC-AUC score of 0.9531.
• It is shared under a permissive CC-BY-4.0 license, allowing for wide reuse and modification.
• The description explicitly defines key metrics like Train ROC-AUC and Val ROC-AUC, aiding interpretation.

Limitations

• The dataset is very small at 9.8 KB, indicating a limited scope of results.
• Column-level documentation is absent; field semantics must be inferred from the description.
• Row count is unknown, which may limit suitability assessment for statistical analysis.

Provenance

Source

figshare

Collection Method

Likely generated from experiments training a deep learning model for medical image analysis.

Freshness

Last updated 2026-05-14 17:44:24; freshness should be verified.