Precision and recall

Two useful measures for understanding what a classifier includes and what it misses.

Google’s Machine Learning Crash Course explains precision and recall with straightforward classification examples. Precision asks how many selected items were actually relevant. Recall asks how many relevant items the system managed to find. Looking at both helps reveal the trade-off hidden by a single accuracy number.

These measures only become meaningful after the task and reference set are defined. A monitoring queue may tolerate extra leads to avoid missing an important claim, while an automatic public action may demand far fewer false positives. We use this resource as an accessible introduction, then attach the measures to a specific workflow rather than treating them as a universal product score.