202412111015
Status: #idea
Tags:
State: #nascient
Precision
Measure of the ratio of correct positive predictions among all the predictions made.
Precision can be remembered mnemonically as the Positive,Parsimonious,Precise side. As opposed to recall which only cares about how likely we are to detect a positive case, precision cares how parsimonious we were, in other words, for all the examples we labeled as positive how many of those actively were. It is computed therefore as:
It is easy to see, how intuitively it competes with its sibling Recall (Statistical Power), as the more confident we are that we captured everything that could be positive, the more likely we are to have been over-eager in our labeling.
This is why in cases where both are important we often use the harmonic mean of the two, that is the F1-Score.