A false negative occurs when a classifier predicts that something is not part of the desired class even though it is. In nodule detection, a missed nodule is a false negative; for a guard dog, a robbery that goes unnoticed is one. section 14.2
False negatives connect classification outcomes to evaluation metrics:
- Contrast a false negative with a true positive, which correctly identifies an item of interest, and a true negative, which correctly rejects an uninteresting item. section 14.2
- Minimize false negatives to improve recall, accepting that a threshold that catches nearly every positive event can also create many false positives. section 14.3.1
- Calculate the false-negative count by subtracting correctly classified positive samples from the total number of positive samples. section 14.3.3
Listing 14.1 Computing counts for true/false positives and negatives
neg_count = int(negLabel_mask.sum()) pos_count = int(posLabel_mask.sum()) trueNeg_count = neg_correct = int((negLabel_mask & negPred_mask).sum()) truePos_count = pos_correct = int((posLabel_mask & posPred_mask).sum()) falseNeg_count = pos_count - pos_correct falsePos_count = neg_count - neg_correct
Inspect the classification-count calculation, including the positive-sample total, correctly predicted positives, and resulting false-negative count. section 14.3.3
