Recall Metric

A Recall Metric is a performance metric for a binary classification model that is based on the Probability that a true test instance is a positive prediction

• AKA: Recall, True Positive Rate, Sensitivity, R, TPR, Positive Predictive Value.
• Context:
  • It can be Estimated by: TP / (TP + FN)
    • # of correct answers given by the system as a proportion of the total # of possible correct predictions)
    • the proportion of cases with a positive test result who are correctly diagnosed.
  • It can be illustrated over a series of cutoffs for defining an Accurate Prediction with a Receiver Operator Curve
• Example(s):
  • The probability of a positive test result in a patient who has the disease under consideration. E.g. probability that a test for cancer will predict that a patient has cancer when in fact they do have cancer.
• Counter-Example(s):
  • a Precision Metric.
  • an F-Measure.
• See: True Negative Rate; False Positive Rate; Receiver Operator Curve; Prevalence; Sensitivity.

---

---

References

2011

• (Sammut & Webb, 2011) ⇒ Claude Sammut. Geoffrey I. Webb. (2011). "Recall." In: (Sammut & Webb, 2011) p.829
• (Sammut & Webb, 2011) ⇒ Claude Sammut. Geoffrey I. Webb. (2011). "Sensitivity." In: (Sammut & Webb, 2011) p.901

2009

• (Wikipedia, 2009) ⇒ http://en.wikipedia.org/wiki/Sensitivity_(tests)#Sensitivity
  • A sensitivity of 100% means that the test recognizes all sick people as such. Thus in a high sensitivity test, a negative result is used to rule out the disease.
  • Sensitivity alone does not tell us how well the test predicts other classes (that is, about the negative cases). In the binary classification, as illustrated above, this is the corresponding specificity test, or equivalently, the sensitivity for the other classes.
• http://www.health.state.mn.us/newbornscreening/glossary.html#positive
  • Positive predictive value - This is a measure of how well a test correctly finds individuals who truly have the condition being checked for. It is the proportion of individuals with positive test results who are correctly diagnosed.

2000

• 2000_SpeechAndLanguageProcessing
  • "Recall is a measure of how much relevant information the system has extracted from the text; it is thus a measure of the coverage of the system."

1998

• (Kohavi & Provost, 1998) ⇒ Ron Kohavi, and Foster Provost. (1998). "Glossary of Terms." In: Machine Leanring 30(2-3).
  • True positive rate (Recall, Sensitivity): d/(c+d).
  • Sensitivity: True positive rate (see Confusion matrix).

---