Neural Network Lexicon

Short Definition

Evaluation metrics measure model performance.

Definition

Evaluation metrics quantify how well a model performs on unseen data. Unlike loss functions, metrics are often task-specific and designed for human interpretation.

Choosing the right metric is crucial for meaningful evaluation.

Why It Matters

A model can optimize loss while performing poorly on the real objective.

How It Works (Conceptually)

• Select task-appropriate metrics
• Evaluate on validation or test data
• Compare models fairly

Minimal Python Example

metric = correct_predictions / total

Common Pitfalls

• Using the wrong metric
• Evaluating on training data
• Ignoring class imbalance

Related Concepts

• Accuracy vs Loss
• Model Evaluation
• Generalization