A true negative (TN) is a test example whose true class is negative and which the classifier correctly predicted as negative. It’s one of the four entries of the Confusion matrix for binary classification, alongside True positive (TP), False positive (FP), and False negative (FN).
In context: if the positive class is patient has heart disease, a true negative is a healthy patient correctly classified as healthy. If the positive class is email is spam, a true negative is a legitimate email correctly delivered to the inbox.
True negatives are the “good” predictions on the negative side of the confusion matrix, complementary to true positives. Together, TN and TP make up the main diagonal — the cases the classifier got right. The classifier’s accuracy is .
The metric most directly tied to TN is specificity (also called true negative rate, TNR):
Of all the actual negatives, what fraction the classifier correctly identified as negative. High specificity means the model rarely raises false alarms.
True negatives often dominate the counts when the dataset is imbalanced with many negatives and few positives. In that regime, a trivial classifier that always predicts negative gets a high TN count and a high accuracy by doing nothing useful — and this is exactly why accuracy alone is misleading for imbalanced problems, and why we look at recall and precision separately.
The classification context determines which class is positive and which is negative. In medical screening, the diseased class is typically positive (we want to catch it); the healthy class is negative. Swap the labels and TN swaps with TP, FP with FN — same data, mirrored confusion matrix.