What are true negatives?
True negatives are the cases a classifier labels as negative that really are negative. A phishing filter that passes a genuine client email, a credit model that approves a borrower who then repays, and an inspection camera that accepts a good part have each produced a true negative. The count sits in one cell of the confusion matrix and drives two metrics: specificity, which is true negatives divided by all actual negatives, and the false positive rate, which is its complement. Accuracy also counts true negatives as correct answers, which is where accuracy becomes misleading.
Why true negatives matter
In most business problems negatives vastly outnumber positives, so true negatives are the biggest number in the matrix. At an accounting firm, a phishing filter might see 50,000 legitimate emails for every 50 attacks. A filter that passes everything scores 50,000 true negatives, zero false positives, and accuracy above 0.99 while catching no attacks at all. The true negative count inflates accuracy without saying anything about the cases that matter.
True negatives still carry real value. Each one is a legitimate email delivered, a good customer approved, or a sound part shipped without a manual check. When a model’s threshold is tightened to catch more positives, true negatives are what get converted into false positives, and the people affected are usually customers or colleagues doing nothing wrong. Reporting true negatives alongside the other three cells keeps that cost visible when thresholds are debated.