Question

You deployed a credit-risk model to a managed online endpoint in Azure Machine Learning. Business stakeholders report that predictions have become less reliable over the past two months. You configure an Azure Machine Learning model monitor to compare recent production inference data against the training dataset used as the baseline. You want the monitor to specifically identify when the statistical distribution of incoming feature values diverges from the baseline distribution and to raise an alert when a configurable threshold is exceeded. Which monitoring signal should you configure?

Accepted Answer

Data drift signal comparing production feature distributions to the baseline dataset

Answer

Prediction drift signal comparing the distribution of model outputs over time

Answer

Data quality signal checking for null rates, type mismatches, and out-of-bounds values

Answer

Feature attribution drift signal comparing feature importance rankings