What reference values of ROC AUC are considered good or bad?

ROC AUC (Area Under the Receiver Operating Characteristic Curve) is a metric of binary classification quality, indicating the model's ability to distinguish between classes.

• A value of 0.5 corresponds to random guessing — the model is no better than random choice.
• Values closer to 1.0 indicate a high ability of the model to correctly classify objects.

General guidelines:

• 0.5–0.6 — poor quality, almost random.
• 0.6–0.7 — weak model.
• 0.7–0.8 — satisfactory, acceptable.
• 0.8–0.9 — good model.
• 0.9–1.0 — excellent model.

However, these thresholds depend on the task and data. In some fields (e.g., medicine), even 0.75 may be considered a good result, while in others — 0.85 may be insufficient.