Skip to main content

Validation Metrics

The Validation Metrics page provides detailed insights into model performance, dataset composition, and evaluation statistics for completed validation runs. It is designed for deep inspection of prediction quality, offering both numerical and visual breakdowns of results.

1. Navigation and Tabs​

From the breadcrumb Home > Validation > Validation Metrics, users can switch between:

  • Summary – Overview of all validation runs
  • Validation Metrics – Detailed performance analysis
  • Start Validation – Launch a new validation
  • External Model Validation – For models not trained on the platform
  • Archived Validations – Previously stored results
  • Exported Metrics – All exported analysis files

2. Key Sections​

The page is organized into three main components:

A. Model Card​

Displays metadata related to the evaluated model:

  • Model Name, ID, and Type
  • Framework and Version
  • Tags describing modality, use case, etc.
  • Linked Models, if any

B. Data Card​

Summarizes the dataset used in validation:

  • Dataset Name and Source
  • Dataset Size and Modality
  • Annotation Details (source and version)

C. Validation Metrics Panel​

Includes comprehensive metrics:

  • Accuracy, Precision, Sensitivity, Specificity
  • F1 Score (Macro and Micro)
  • AUC and NPV (Negative Predictive Value)
  • True/False Positive and Negative Counts

3. Visual Performance Analysis​

The following visual tools are included:

  • ROC Curve
  • Precision-Recall Curve
  • Confusion Matrix
  • Loss Samples Viewer (top incorrect predictions)
  • Class-wise Performance Breakdown
  • Confidence Threshold Performance
  • Confidence Distribution Graphs
  • Training vs. Validation Metrics
  • Population Distribution
  • Most Confused Classes
  • Lift Chart (if supported)

These tools help users visually interpret where and how model predictions deviate from ground truth.

4. Actions and Tools​

A dropdown menu in the top-right corner enables:

  • Exporting Metrics and performance tables
  • Editing View preferences
  • Creating Subcohorts for focused evaluation
  • Generating Subcohort Analysis based on custom filters

Note: The Validation Metrics page is your go-to tool for understanding how your model behaves across different classes, thresholds, and datasets. Use it to verify performance, discover edge cases, and guide future improvements.