Model Evaluation ROC Space

Prediction Space

Test samples 44

Class separation 55%

Decision threshold 0.5

Noise level 0.4

True positives

False positives

Precision

Recall

How it works

Every classifier's real performance is a whole curve, not one number — moving the decision threshold always trades precision against recall.

precision = TP/(TP+FP)
recall = TP/(TP+FN)

Test samples — held-out labelled samples the model is scored against.
Class separation — how well-separated the model's predicted scores are between the two true classes.
Decision threshold — score cutoff above which a prediction counts as positive.
Noise level — label/measurement noise blurring the separation between classes.

Choosing this threshold is a real deployment decision — a cancer screen and a spam filter want very different points on the same precision-recall curve.

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.