AI Dictionary › Fondamenti AI

Evaluation metrics

Metriche di valutazione

Evaluation metrics are the numeric measures used to quantify how well a machine learning model performs its task, allowing different models to be compared, the best one to be chosen, and a decision made on whether the achieved performance is sufficient for real-world use. Without a shared metric, claiming a model works well is a vague, unverifiable statement.

Definition

How it works

The choice of metric depends on the type of problem. For regression, measures like mean squared error or mean absolute error are used, quantifying how far predictions deviate from actual values. For classification, accuracy, precision, recall, F1-score and area under the ROC curve are used, each sensitive to a different aspect of the errors made. There is no universally best metric: the choice depends on what truly matters for the specific problem, for example whether a false positive or a false negative is worse.

Applications

In applied AI, metrics guide every technical decision: which model to put into production, when an update actually represents an improvement, when a system has degraded over time and needs retraining. They are also a communication tool between technical teams and business decision-makers, who rarely understand algorithmic details but can grasp that precision rose from 80 to 90 percent.

History & etymology

The systematic use of quantitative measures to evaluate the quality of statistical predictions belongs to early twentieth-century statistics, but the formalization of the metrics standard in machine learning today took shape starting in the 1960s and 70s, in fields like pattern recognition and information theory, before spreading widely with the discipline's growth in the 1990s and 2000s.

How it's used in Grace

Grace's 7 evaluation dimensions and 0-100 score are, in effect, evaluation metrics applied to the quality of your prompts: the same principles of objective measurement used to evaluate a machine learning model.

Related terms

• Confusion matrix
• Precision and Recall
• ROC curve and AUC
• Cross-validation
• Classification

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.