HumachLearnLearn · explore · apply
Start learning
Working playground

Metrics & Threshold Playground

Manipulate class separation, prevalence and the decision threshold. See the confusion matrix and operational metrics respond immediately.

Lab concept guide

What to observe while you experiment

A decision threshold converts probabilities/scores into class labels, so operational metrics are properties of both model scores and the chosen threshold. Prevalence and class separation alter confusion counts and can change which metric is useful.

MechanismMove the threshold and trace how individual predictions cross it, changing TP/TN/FP/FN and therefore precision, recall, specificity and related curves.
Failure modeTreating 0.5 as universally optimal or comparing threshold-dependent metrics without stating prevalence and decision costs.
VerificationPick one threshold, count the four confusion cells manually, and recompute at least precision and recall; then predict the effect of moving the threshold.
Experiment deliberately
Lower the threshold gradually. Before each move, predict the direction of recall, false positives and precision; verify against the confusion matrix.
Ready.