Confusion Matrix

A confusion matrix is a table that breaks down a classification model's predictions into four categories, showing not just whether the model is right or wrong, but exactly how it is wrong. It is one of the most useful tools for understanding model performance.

The four cells (binary classification)

For a model that predicts "yes" or "no" (such as spam detection):

• True Positives (TP) — model said "spam" and it was spam. Correct.
• True Negatives (TN) — model said "not spam" and it was not spam. Correct.
• False Positives (FP) — model said "spam" but it was not spam. A false alarm.
• False Negatives (FN) — model said "not spam" but it was spam. A missed detection.

Why the breakdown matters

Accuracy alone tells you the overall percentage correct, but the confusion matrix tells you the story behind that number. Two models with identical accuracy can have very different error patterns:

• Model A catches ninety-five per cent of spam but sends ten per cent of legitimate emails to the spam folder
• Model B catches seventy per cent of spam but almost never misclassifies legitimate email

Which is better depends on your use case. The confusion matrix makes this trade-off visible.

Metrics derived from the confusion matrix

• Precision — of all the items the model labelled positive, what fraction actually were? (TP / (TP + FP))
• Recall — of all the actual positives, what fraction did the model catch? (TP / (TP + FN))
• F1 Score — the harmonic mean of precision and recall, balancing both

Multi-class confusion matrices

For models with more than two classes, the matrix expands. A three-class model produces a three-by-three grid. This reveals which specific classes the model confuses with each other — invaluable for understanding failure patterns.