Recall
metric measuring how much actual positive samples are located correctly by the model.
Sensitivity (recall) is a metric that measures how many examples that are actually positive are correctly detected by the model, and real positives are calculated by division of false negatives with real positives. Different from certainty, remember, focuses on the question of how many of the actual positive cases of the model capture without missing, so it is also called sensitivity (sensitivity) or real positive rate. High sensitivity means that the positive samples of the model are low the miss rate.
Sensitivity is particularly critical in areas where a positive case has severe consequences: classification of a cancer case in medical diagnosis as "healthy" is much more dangerous than the fact that healthy one is directed to additional tests on unnecessary location, so high recall is targeted in screening tests. The cost of real threats that have been missed in similar fraud detection or security systems is high. However, trying to maximize the sensitivity, the more examples of the model push positively to mark, which can lead to the drop of certainty, so in practice the preciseness-weight balance according to the requirements of the application, the decision threshold is set by adjusting.
Sensitivity is particularly critical in areas where a positive case has severe consequences: classification of a cancer case in medical diagnosis as "healthy" is much more dangerous than the fact that healthy one is directed to additional tests on unnecessary location, so high recall is targeted in screening tests. The cost of real threats that have been missed in similar fraud detection or security systems is high. However, trying to maximize the sensitivity, the more examples of the model push positively to mark, which can lead to the drop of certainty, so in practice the preciseness-weight balance according to the requirements of the application, the decision threshold is set by adjusting.
Member comments