What does recall measure in a machine learning model?

Recall, also known as sensitivity or true positive rate, specifically measures the ability of a model to identify all relevant instances of a particular class. In the context of a binary classification problem, it is calculated by taking the number of true positives (the instances correctly identified as positive) and dividing it by the total number of actual positives in the dataset. This means that recall reflects how well the model can capture the positive cases among all the actual positive instances present.

Therefore, when stating that recall is the percentage of positives found compared to all relevant instances, it encompasses the essential characteristic of this metric: that it focuses on the effectiveness of the model in recognizing positive cases, emphasizing its importance in scenarios where identifying positives is crucial, such as in medical diagnoses or fraud detection.

In terms of the other options, while true positives are indeed part of the definition (as seen in the first option), recalling that recall considers all positive instances makes choice C the most accurate definition. Meanwhile, negative classifications and overall accuracy pertain to different aspects of model evaluation, such as specificity and general performance metrics, respectively, which do not focus on the identification of positives.