AI Dictionary › Fondamenti AI

Precision and Recall

Precisione e Richiamo (Precision-Recall)

Precision and recall are two complementary metrics for evaluating a classification model, each sensitive to a different type of error. Precision answers the question: among all the cases the model flagged as positive, how many actually were? Recall answers a different question: among all the truly positive cases, how many did the model manage to find?

Definition

How it works

Precision is calculated by dividing true positives by the sum of true positives and false positives: a model with high precision makes few wrong calls, but might miss many real ones. Recall is calculated by dividing true positives by the sum of true positives and false negatives: a model with high recall finds nearly all positive cases, but may do so at the cost of flagging many that were not positive. The two measures are often in tension: increasing one tends to decrease the other, and the model's decision threshold lets you shift the balance one way or the other. F1-score combines the two into a single number by computing their harmonic mean.

Applications

Whether to favor precision or recall depends on the relative cost of the two types of error: a fraud detection system often favors high recall, even at the cost of some false alarms, because missing a real fraud is more costly; a system that automatically filters content for removal often favors high precision, to avoid unfairly penalizing legitimate content. These are central metrics in information retrieval, medical diagnostics, content moderation and anomaly detection.

History & etymology

The two concepts originate in distinct fields that later merged: precision and recall were formalized in the information retrieval community, the search for relevant documents in large archives, as early as the 1950s and 60s, to measure how well a search system returned relevant results without missing too many. With the growth of machine learning and automatic classification, these two indicators migrated permanently into the standard vocabulary of model evaluation.

Related terms

• Confusion matrix
• ROC curve and AUC
• Classification
• Imbalanced data
• Evaluation metrics

More in Fondamenti AI

• LLM (Large Language Model)
• Token
• Context Window
• Temperature
• AI Hallucination
• Fine-tuning
• RAG (Retrieval Augmented Generation)
• Embedding

Put it into practice

• Real scenario: apply this concept with an instant AI verdict.
• Daily gym: short drills to train dictionary concepts.
• Playoff game: challenge other players, elimination-style.

📱 Download the Android app (beta) iOS coming soon

Say what you mean. Get what you need.

Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.