Precision and Recall Explainer
Precision and recall describe different aspects of a classifier’s positive predictions. Precision is the share of predicted positives that are truly positive. Recall is the share of actual positives that the model successfully identifies. The two metrics answer different operational questions.
For a spam filter, high precision means messages marked as spam are usually spam, limiting false alarms. High recall means the filter catches most spam, limiting misses. Changing the decision threshold often improves one while worsening the other because the model becomes more or less willing to label cases positive.
Class prevalence affects interpretation. A model can have high accuracy while performing poorly on a rare class, and precision can fall when the positive condition becomes less common. Metrics should be calculated on data resembling the population and decision setting where the model will be used.
The right balance is a policy and product choice, not a universal mathematical answer. Medical screening, fraud review, search retrieval, and moderation attach different costs to false positives and false negatives. Teams may use precision-recall curves, F-scores, workload limits, and explicit risk analysis to select a threshold.
Precision asks how many positive predictions were correct, while recall asks how many actual positives were found.
Key terms include precision, recall, positive class, false positive, false negative, threshold, prevalence, precision-recall curve and F-score.
Precision starts from everything the model called positive and asks what fraction was correct. Recall starts from all real positives and asks what fraction was found.
A system that predicts almost nothing positive can avoid many false alarms but may have unusably low recall.
Moving the threshold changes the counts behind both metrics, so teams should inspect performance across a practical range rather than at one arbitrary point.
The selected operating point should account for human review capacity, consequences of misses, consequences of false alarms, and distribution shift.
Explore more "Explainers"
Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.