La moyenne micro est une méthode statistique couramment employée dans classification multi-classes tasks within the field of Intelligence artificielle and Apprentissage automatique. It provides a way to assess the performance of a classification model by considering all classes together, rather than calculating metrics for each class separately. This approach is particularly useful when dealing with jeux de données déséquilibrés où certaines classes peuvent avoir beaucoup plus d’échantillons que d’autres.
In micro-averaging, the true positives, false positives, and false negatives are summed across all classes before calculating the overall precision, recall, or F1 score. This means that each instance contributes equally to the final metric, regardless of which class it belongs to. The formula for micro-average precision (P) and recall (R) can be expressed as:
- Précision micro-moyenne : P = (Somme des Vrais Positifs) / (Somme des Vrais Positifs + Somme des Faux Positifs)
- Rappel micro-moyenne : R = (Somme des Vrais Positifs) / (Somme des Vrais Positifs + Somme des Faux Négatifs)
Micro-average is particularly advantageous in scenarios where the focus is on overall model performance rather than on individual class performance. This can help in providing a more holistic view of a model’s capabilities, especially in applications such as classification d'image, text categorization, and speech recognition.