Dans le contexte de classification tasks in intelligence artificielle and apprentissage automatique, Moyenne Macro is a method used to evaluate the performance of a model across multiple classes. Unlike Moyenne Micro, which aggregates the contributions of all classes to compute the average performance, Macro-Average treats each class equally regardless of its taille.
Pour calculer la Moyenne Macro, vous commencez par calculer la métrique d’évaluation (such as precision, recall, or F1-score) for each class individually. Then, you take the arithmetic mean of these scores across all classes. This approach ensures that each class has an equal weight in the final average, which can be particularly important in scenarios where class distributions are imbalanced.
For instance, consider a scenario with three classes: Class A (100 instances), Class B (10 instances), and Class C (5 instances). A model may perform exceedingly well on Class A and poorly on Classes B and C. If you only look at précision globale, it might seem like the model is performing well. However, Macro-Average will highlight the poor performance on the minority classes, providing a more balanced view of the model’s effectiveness.
Alors que la Moyenne Macro est utile pour comprendre la performance du modèle in multi-class settings, it is essential to consider it alongside other metrics, particularly when dealing with imbalanced datasets, to get a comprehensive view of a model’s performance.