MLA-C01 Question 378
Select 2You are managing a machine learning model deployed on Amazon SageMaker that predicts customer churn for a subscription service. Over time, you notice a decrease in the model's prediction accuracy. Upon further investigation, you find that the customer behavior patterns have shifted due to a recent change in subscription pricing. What would be the most appropriate actions to address this issue?
- A
Monitor the model's data distribution using Amazon SageMaker Model Monitor to detect data drift.
- B
Retrain the model with a new dataset that reflects the recent customer behavior changes.
- C
Increase the instance size of the SageMaker endpoint hosting the model to improve prediction speed.
- D
Use Amazon SageMaker Clarify to identify potential bias introduced by the new subscription pricing.
- E
Adjust hyperparameters of the existing model to compensate for the observed drift.
Show answer and explanation
Correct answers: A, B
Explanation
The decrease in model accuracy is caused by data drift, where the input data distribution has changed due to shifting customer behavior. To address this, you should first monitor the data for drift using Amazon SageMaker Model Monitor and then retrain the model with a new dataset reflecting the updated behavior. Simply optimizing infrastructure or hyperparameters will not address the underlying issue of drift, and SageMaker Clarify is not designed to resolve this problem.
- A. Correct.
Monitoring the model's data distribution with SageMaker Model Monitor is a key step in detecting data drift and understanding how the input data has changed.
- B. Correct.
Retraining the model with a dataset that includes updated customer behavior ensures that the model adapts to the new patterns introduced by the subscription pricing change.
- C. Incorrect.
Increasing the instance size of the SageMaker endpoint improves performance but does not address the issue of decreased prediction accuracy caused by data drift.
- D. Incorrect.
SageMaker Clarify is used for bias detection and explainability, which is not directly relevant to addressing the problem of data drift in this scenario.
- E. Incorrect.
Adjusting hyperparameters without addressing the underlying data distribution change will not effectively resolve the issue caused by drift.