Classification Accuracy

Classification accuracy measures how often a classifier is right overall. It adds the cases it got correct: true positives and true negatives, then divides by all predictions made, so the result is the share of predictions that matched the true labels. This works because every prediction falls into exactly one of four categories: correct positive, correct negative, false positive, or false negative.

Use accuracy when you want a quick summary of overall performance for a classification model, especially when the classes are fairly balanced. It is common in tasks like spam detection, disease screening, image labeling, and any binary or multiclass prediction problem where you care about the fraction correct.

• Using accuracy alone even when one class is very rare, which can hide poor performance on the minority class
• Swapping false positives and false negatives when counting the confusion matrix
• Forgetting that the denominator is the total number of predictions, not just the correct ones