Recall

Recall measures how many of the objects (e.g. images or point clouds) that should be correctly predicted by an AI’model have actually been correctly predicted. Example: If an AI model predicts whether a part is OK or NOK, the recall for NOK is the proportion of NOK parts that were correctly identified as NOK. In this case, the recall therefore provides an answer to the question: “Of all NOK parts, how many has the AI model detected?” Another important term in this context is precision. Precision indicates how many of the tasks correctly predicted by an AI model are actually correct. For example, if an AI model predicts whether a part is OK or NOK, the precision for NOK is the proportion of parts that were predicted to be NOK and are actually NOK, so the answer is: “If the AI model says NOK, how often is it correct?”