How-to assess predicted categories against known gold standards

Align ground truth data and predictions with Firetail 12.2

With Firetail there is now a fast and easy way to compare predicted categories and known gold standards. This allows you to compute the amount of overlap against reference data and estimate agreements across categories or layers of your behavioral assignments.

Steps

1. Predict from acceleration data (FireSOM module) or overlay external annotations

2. Import CSV annotations

Load external resources, like BORIS video annotation, 3rd party predictions or manually curated standards.

3. Open the annotation evaluation module

4. Link each reference layer (or category) to one or more predicted layers

5. Compile your results

Sort by precision, recall or F-measure. This tool is also a great way to:

• match the most likely ethogram states to abstract predictions or
• compare multiple predictive algorithms or feature sets for similarity