What accuracy metric is standard for evaluating binary classification models?

For evaluating binary classification models, the standard accuracy metric is often referred to simply as accuracy, which measures the proportion of correct predictions made by the model out of all predictions. This metric is expressed as the number of true positives and true negatives divided by the total number of observations.

When you assess a binary classification model, particularly in a scenario where the classes are imbalanced, relying solely on accuracy can be misleading, as it may not reflect the model's performance effectively. Precision, Recall, and Area Under the Curve (AUC) are also important metrics to consider in specific contexts, but accuracy provides a straightforward initial assessment of how often the model is correct across all classes.

Accuracy becomes particularly relevant when the class distribution is relatively balanced, as it captures the direct success rate of the classification task. The value of AUC lies in its ability to measure the performance across all classification thresholds, enhancing the interpretation of a model's predictive power, especially in imbalanced datasets.

In summary, while AUC is a valuable metric for evaluating the overall ability of a binary classifier to discriminate between classes, the standard metric for directly evaluating the accuracy of a binary classification model is simply accuracy itself.