Confusion Matrix

A table of predicted versus actual classes that breaks down a classifier's correct and incorrect calls.

Definition

A confusion matrix tabulates a classifier's predictions against the true labels. For binary classification it has four cells: true positives, true negatives, false positives, and false negatives.

Extending to more than two classes, the matrix becomes a grid whose off-diagonal cells show exactly which classes are mistaken for which. This reveals systematic confusions, such as two visually similar categories, that a single accuracy figure would hide entirely.

Normalizing the matrix by row or column turns raw counts into rates, recall along one axis and precision along the other, which makes performance comparable across classes of very different sizes. For multi-class problems, reading the matrix reveals not just how often the model errs but its characteristic mistakes, information that guides targeted data collection or class-specific fixes far better than a single accuracy figure.

Predicted vs Actual (binary)

From these counts flow the standard metrics: accuracy, precision and recall, and specificity.

Why it matters

A single accuracy number hides where a model fails. The confusion matrix shows whether errors are false alarms or misses, a distinction that matters enormously when the two kinds of error carry different costs, as in safety or medical screening.

Fusion connection

When a classifier flags Hyperion configurations as safe or unstable, the confusion matrix reveals whether it errs toward false alarms or dangerous misses, guiding how conservatively its output is used.