Often algorithms give some sort of confidence measure for making a desicison, for example, a cancer diagnois. We can then make the decision based on a threshold value, making a postive diagnosis if the confidence exceeds the threshold.. If the threshold of confidence is very high, we ensure high precision, that is if we make diagnosis, then it is almost certainyl right; but this then means that we are likley to miss some potential cases (false positives), that is low recall. Alternateively a low threshold value will give high {[recall}}, we are unlikely to miss positive cases, but more likely to falsely diagnose someone (false negative), that is low precision. In general, there is a trade-off between precision and recall and the correct balance is a cost–benefit trade-off depending on the relative impact of false positives and false negatives.
Defined on page 180
Used on pages 180, 193