Accuracy
FoundationsCore AI Concepts
Accuracy is the proportion of correct predictions a model makes relative to the total number of predictions evaluated. In machine learning it is the easiest metric to read and one of the easiest to be misled by. On an imbalanced dataset, a model that simply predicts the majority class scores well while failing to identify the rare events that matter, such as fraudulent financial transactions or equipment failures.
In practice
Accuracy on its own is not a validation result. Ask for precision and recall alongside it wherever the thing being detected is rare, which covers most fraud, credit and failure-prediction work. A vendor quoting a single accuracy figure for an imbalanced problem is quoting the least informative number available.