When the examples of our data set contain a very large number of one label (y=0) [e.g. the patient does NOT have cancer] and a comparably smaller number of the other label (y=1) [e.g. the patient has cancer], the term “skewed classes” is often used to describe such a data set.
In the case of skewed classes it is common to use the metrics of “precision” and “recall” to measure how good our model is at making correct predictions.
Where “precision” represents the ratio of “true positives” to “predicted positives” (true positives / (true positives + false positives) and “recall” represents the ratio of “true positives” to “actual positives” (true positives / (true positives + false negatives)
Ideally we want both of these numbers to be high.
Using the term “precision” to represent the ratio of “true positives” to “predicted positives” kind of makes sense because our motivation is to avoid telling someone that they have a condition when they do not have it. So we want to “be precise” in our prediction.
However, in the case of “recall” our motivation is to avoid not telling someone that they have a condition when they do indeed have it. I don’t understand how this concept is in any way related to the english word “recall”.
