Engineering Sandbox rollout · bundle-mounted ML essays
Which mistakes change when you lower the threshold?
Lower the cutoff and count newly predicted positives in the confusion matrix. Recall counts recovered positives; precision asks how many positive predictions are right. F1 combines these ratios but does not encode the real cost of a miss or false alarm.
How to use this page
Move the threshold in small steps and track true positives, false positives, and false negatives before reading the ratios. Lowering it cannot remove a predicted positive for fixed scores, but precision need not change monotonically as examples cross it.
Loading Precision and Recall lesson…