Skip to main content
All terms
Evaluation

Accuracy

The share of a model's predictions that are correct against known answers.

Definition

Accuracy is the fraction of predictions a model gets right, measured against ground-truth answers. It is simple and intuitive, which makes it a default metric for classification and multiple-choice tasks. On imbalanced data it can mislead, since a model can score high by always predicting the majority class, so it is often reported alongside precision, recall, and F1.