Rappel, also known as sensitivity or true positive rate, is a key performance metric used to evaluate the effectiveness of a classification model in the domaine de l'intelligence artificielle and apprentissage automatique. It quantifies the ability of a model to correctly identify all relevant instances within a dataset.
Mathématiquement, le rappel est défini comme le rapport entre les vrais positifs et le nombre total d'instances positives réelles. La formule du rappel est :
Rappel = Vrais Positifs / (Vrais Positifs + Faux Négatifs)
Dans cette formule :
- Vrais Positifs (VP) désignent les instances correctement classées comme positives.
- Faux Négatifs (FN) are the instances that are incorrectly classified as negative, despite being positive.
Recall is particularly important in scenarios where the cost of missing a positive instance is high, such as in medical diagnoses or détection de fraude. A high recall score indicates that the model is effective at capturing most of the relevant instances, while a low score suggests that many positive instances are being overlooked.
However, it is essential to consider recall in conjunction with other metrics, such as precision (the ratio of true positives to the total predicted positives) and F1 score (the moyenne harmonique of precision and recall), to get a comprehensive understanding of a model’s performance. Balancing recall and precision is crucial, as focusing solely on maximizing recall may lead to a high number of false positives.