I know the definition of true positive rate and true negative rate, I know how to calculate them, and I know the mnemonic “SpIn and SnOut” but I need an analogy or something because I don’t truly understand the difference.
Let’s say a compound fracture has a bone sticking out 100% of the time. Would “bone sticking out” be 100% specific or 100% sensitive to a compound fracture?
In: 2
A classifier that is “specific” is really picky. If even the most pedantic detail doesn’t line up with what the classifier expects, it will assign a negative classification. As a result, the set of objects which are assigned positive classifications will be extremely pure, where the “impurities” are false positives.
A classifier that is “sensitive” is really good at noticing not-so-obvious details. When presented with an ambiguous object to classify, it is sensitive enough to pick up on subtle hints of the true class of the object. So, of all the true positives this classifier is shown, it will accept a lot of them.
These two concepts are independent, it is possible for a classifier to have any level of sensitivity and specificity. However, they are offen correlated. If you have a very sensitive classifier, it’s a lot more likely to be “tricked” into accepting true negatives. So in general, increasing sensitivity will reduce specificity.
Latest Answers