In a fraud detection evaluation, precision is the number of correctly flagged records divided by all flagged records. It answers how often an alert is right on the evaluated data.
Precision depends on which cases the system flags, which cases appear in the evaluation and how outcomes are labeled. It is often read alongside recall to understand which cases the system misses.
ELI5
For example, if a tool flags 10 invoices and 8 have the problem it was looking for, its precision is 8 out of 10, or 80 percent.
You would also want to know whether many other problem invoices went unflagged. That is what recall helps measure.
Is precision the same as accuracy?
Precision focuses on predicted positive cases. Accuracy measures the share of all evaluated cases classified correctly.
Can a system have high precision and low recall?
Yes. A conservative system may flag only a few obvious cases and get most of those right while missing many other positives.

