ถอดรหัสเมทริกซ์ความสับสน

For classifiers, one accuracy score hides too much. The confusion matrix lays out exactly which predictions were right and which were wrong. 🧩

The matrix is a small grid: rows are the true classes, columns are the predicted ones. Each cell counts how often that pairing happened.

A true positive is when the model says yes and the answer really is yes. These are the wins you want to maximize.

A true negative is when the model says no and the truth is also no. The model correctly left this case alone.

A false positive is a false alarm: the model predicted yes, but the truth was no. Think of flagging a healthy patient as sick.

A false negative is a miss: the model said no while the truth was yes. Picture letting a real fraud slip through unnoticed.

Notice the two error types differ in cost. A false positive and a false negative are rarely equal, so the matrix keeps them separate.

Accuracy is just the correct cells over the total. It is the diagonal divided by everything, which is why it can mask rare errors.

The most useful cells sit off the diagonal: they show where the model confuses one class for another, pointing to its weak spots.

For many classes, plot the matrix as a heatmap. Bright off-diagonal squares instantly reveal which labels get mixed up most.

Almost every classification metric is built from these four counts. Master the confusion matrix and the rest follow naturally.

Let's make sure each cell of the matrix is crystal clear.

The confusion matrix splits results into true and false, positive and negative. Those four counts power every classification metric. 🎯