Sensitivity (True positive) Link to heading

= Positive among all true

Sensitivity is for describing how sensitive a classifier or test is. Given a “true”, the probability of spotting this “true” is called sensitivity. $$ \text{Sensitivity}=P(\text{positive}|\text{true}) $$ For example, a blood screening is used for spotting a disease. Sensitivity of the blood screening means the probability of getting a positive result when that person is actually sick. A very sensitive blood test will spot almost all sick people and even take healthy people as sick (low specificity).

Specificity (True negative) Link to heading

= negative among all false

Specificity is less straightforward to interpret. It is for describing how specific a classifier is. High specificity usually means a test or classifier is tolerant.

If a classifier is very specific, it means this classifier is giving negative result blindly. It only gives positive result when the objective satisfies very SPECIFIC conditions.

A super specific clinic test will not take a single healthy person as sick, and it will even fail to spot a diseased person (low sensitivity).

If a classifier or test gives positive blindly, as mentioned before, it has high sensitivity but low specificity.

False positive rate Link to heading

= Positive among all false

False positive rate = $P(positive|false)=1-specificity$

False negative rate Link to heading

=Negative among all true

False negative rate = $P(negative|true)=1-sensitivity$