You have deployed a customer churn prediction model to a Vertex AI endpoint. After several weeks, the model's performance, as measured by the AUC, has significantly declined compared to its training evaluation. What steps should you take to diagnose this issue?
