Plot performance across classification thresholds

Sensitivity, specificity, and TSS as the probability cutoff moves from 0 to 1, with the TSS-maximising cutoff marked.

Usage

plot_threshold_performance(predictions, path = NULL)

Arguments

predictions

the predictions element of a fit_patch_model() result

path

where to write a PNG; NULL returns the plot instead

Value

the plot object, or path invisibly when written to disk

Details

The metrics table reports sensitivity and specificity at the default 0.5 cut, which is rarely the right one for imbalanced data — the original pipeline binarised its projections at a ROC-derived cutoff for exactly this reason. This shows what the trade-off looks like across the whole range, and where it is best balanced.