liveBook · Manning

Recall

As explained in Fight Fraud with Machine Learning

Recall is the proportion of actual positive cases that are correctly identified.

Recall is the proportion of actual fraudulent cases correctly identified: true positives divided by true positives plus false negatives. It measures how many fraudsters the system catches. section 1.6.3

Why recall matters

A system that flags every transaction or email as fraudulent achieves 100% recall, but it also catches many genuine cases. Aggressive detection improves recall while typically reducing precision. section 1.6.3

Use recall alongside related evaluation concepts:

  • Compare recall with precision: maximizing recall reduces missed fraud but can increase false alarms. section 1.6.3
  • Prefer recall when missing fraud is unacceptable; prefer precision when falsely flagging genuine activity is especially costly. section 1.6.3
  • Use recall and precision instead of accuracy when fraudulent and genuine cases are highly imbalanced. section 1.7
Equation 1.1 Defining f-score as the harmonic mean of precision and recall
F-score = 2 * Precision * Recall / (Precision + Recall)

The F-score combines recall and precision through their harmonic mean when one metric alone is insufficient. section 1.6.3

How Fight Fraud with Machine Learning covers it

How deep it goes
Defines recall from true positives and false negatives, then stops short of implementation or threshold-selection procedure.
Examples
Uses fraudsters caught, fraudulent emails, and flagging every transaction to illustrate recall and its extremes.
What it assumes
Assumes familiarity with confusion-matrix counts and genuine-versus-fraudulent classification cases.
What you can do after
Evaluate fraud-detection systems by emphasizing missed-fraud costs alongside false-positive costs.
How it connects
Connects recall to precision, confusion matrices, F-scores, class imbalance, and precision-recall curves.

Also explained in

Compare how 35 books explain recall →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Fight Fraud with Machine Learning. The section links point to the parts of the book it draws on.