Confusion matrix - understand the sensitivity and specificity
definition
Case
Assuming that a total of 100,000 patients, among them suffering from cancer of 200 people, or 0.2 percent prevalence rate, the actual test results are as follows:
A positive test | Test negative | total | |
---|---|---|---|
The actual positive | 160 | 40 | 200 |
The actual negative | 29940 | 69860 | 99800 |
total | 30100 | 69900 | 1000000 |
Based on the above data:
Namely:
sensitivity (also known as the recall) represents the actual number of cases detected in the proportion of the total number of cases reached 80%, the proportion of patients that can detect real out
specifically indicates that the test without sick people, determining exclude 70% probability of illness, will have 30% chance of being tested positive, but not actually sick. That can not be detected proportion of real illness.
Better model, so that all samples can be distributed in four quadrants proportion more optimized to enhance the sensitivity and specificity, increasing the proportion of the sample positive diagonal, i.e. improved detection.