Recall (also called sensitivity or true positive rate) measures what share of the actual positive cases a model detects: correctly identified positives divided by all real positives. If a fraud filter catches 80 out of 100 real fraud cases, its recall is 0.8. High recall matters when missed cases are costly, such as in medical diagnosis. Because it ignores false alarms, it is usually considered together with precision, for example in the F-score.
The F-score is a metric for evaluating classification models. The common F1 score is the harmonic mean of precision (how many of the predicted hits are correct) and recall (how many of the actual hits were found), and ranges from 0 to 1. It is especially useful with imbalanced data: a spam filter that almost never flags spam can have high accuracy but a low F1 score.
A confusion matrix is a table that compares a classifier's predictions with the actual classes. For two classes, it shows true positives, false positives, true negatives and false negatives. From this, metrics such as precision, recall and F1 score can be calculated, and it becomes clear which errors occur. Example: a spam filter that flags many genuine emails as spam has many false positives, which is often more costly than a missed spam message.
The ROC curve (receiver operating characteristic) shows how well a classification model separates two classes. For every possible decision threshold it plots the true positive rate (recall) against the false positive rate. The closer the curve reaches towards the top-left corner, the better; the diagonal corresponds to pure guessing. The area under it (AUC) summarises the curve in a single number and helps, for example, when comparing different fraud detection models.
Mean average precision (mAP) is a metric that assesses how well a system finds relevant results and ranks them near the top. Average precision is calculated across the result list for each query or object class; mAP is the mean across all of them. It is a standard measure for search engines, recommender systems and object detection in images, where it also depends on the required overlap of bounding boxes.
