Classifier Evaluation Playground
Explore how a binary classifier performs — both discrimination (can it rank positives above negatives?) and calibration (do its probabilities mean what they say?). Simulate a dataset, or paste your own labels and scores.
How far apart the positive and negative score distributions sit. 0 = no signal.
One row per prediction. label is 0 or 1; score is a probability in [0,1] (enables calibration) or any ranking score (discrimination only). CSV, tab, or space separated. A header row is skipped.
Auto matches predicted prevalence to observed. Enable to set manually.
ROC curve
True-positive vs. false-positive rate across all thresholdsPrecision–Recall curve
More informative than ROC when events are rareReliability diagram
Predicted probability vs. observed frequency — on the diagonal = calibratedAt threshold
Set your inputs and press Evaluate to see metrics and curves.