All courses › Machine Learning and Data Analysis › Precision and recall

Precision and recall

Precision answers how often the model is right when it says yes. Recall answers how many of the real cases it found. F1 is a compromise between them. Which matters most depends on what costs more: false alarms or missed cases.

P = \frac{TP}{TP + FP}

precision

R = \frac{TP}{TP + FN}

recall

F_1 = \frac{2PR}{P + R}

F1 score

Symbols

$P$	precision
$R$	recall
$F_1$	harmonic mean of P and R

Example

TP = 40, FP = 10, FN = 20:

$P = 0.8$ , $R \approx 0.67$ and $F_1 \approx 0.73$ .

In cancer screening recall matters most, in a spam filter precision matters most.

Practise classification and evaluation for free →

← The confusion matrix · Sigmoid and threshold →

Part of Machine Learning and Data Analysis: Classification and evaluation.