Machine Learning/Bank Fraud Detection

There are many applications for machine learning and binary classification, and one of these is the detection of illegitimate transactions on ATMs at banks.

• True positive: The action was fraudulent, and the algorithm correctly identified the action as such.
• False positive: The action was not fraudulent, yet the algorithm wrongly identified the action as such. (An innocent user was wrongly denied the service.)
• True negative: The action was not fraudulent, and the algorithm correctly identified the action as being legitimate.
• False negative: The action was a fraud, but the algorithm misclassified the action as legitimate. (A bad-faith user was wrongly granted the service.)

The recall is defined as number of true positives divided by the total number of true positives plus false negatives.