In the process of reading papers, we often encounter situations in which the performance of classifiers is described by specificity and sensitivity. The expressions of these two indicators are somewhat vague, and we will record them after consulting relevant information here.

Consider a binary classification with categories of 1 and 0. We take 1 and 0 as positive and negative categories respectively, so there are four actual classification results, as shown in the following table:

Some other evaluation indicators can be drawn from this table:

Taking the medical screening of diabetic patients as an example, the sensitivity and specificity are illustrated. In this example, we only use the patient's blood sugar level as an indicator to judge whether he has diabetes. The following figure is a statistical chart of blood sugar levels of normal people and diabetic patients:

That is to say, the point V34 in the figure has high sensitivity and high specificity.

Which great god can explain sensitivity and specificity? I have been studying for several years, and I have never been very clear about it. -Zhihu

ROC curve-Baidu encyclopedia

Positive and negative predictive values