MLA-C01 Question 380
Select 3You are a Machine Learning Engineer managing a model deployed on Amazon SageMaker for a fraud detection system. Over time, you notice a decline in the model's performance. Which of the following steps can help you monitor and address both data quality and model performance issues?
- A
Set up Amazon SageMaker Model Monitor to track data drift in the input features.
- B
Use CloudWatch Logs to monitor the latency of model predictions.
- C
Enable Amazon SageMaker Clarify to detect bias in the incoming data.
- D
Periodically retrain the model using new data without monitoring data drift.
- E
Configure Amazon SageMaker Model Monitor to track prediction drift.
Show answer and explanation
Correct answers: A, C, E
Explanation
Monitoring data quality and model performance is essential for maintaining the reliability of deployed machine learning models. Amazon SageMaker Model Monitor can track data drift and prediction drift, while Amazon SageMaker Clarify can identify bias in data. These tools provide a comprehensive approach to identifying and addressing potential issues before they significantly impact model performance. Operational metrics like latency (via CloudWatch Logs) and retraining alone do not directly address these concerns.
- A. Correct.
Amazon SageMaker Model Monitor is specifically designed to track data quality issues such as data drift, which could impact the model's performance over time. This is a necessary step for monitoring input features.
- B. Incorrect.
While CloudWatch Logs can help track operational metrics like latency, it does not address data quality or model performance directly. This is unrelated to monitoring data drift or prediction drift.
- C. Correct.
Amazon SageMaker Clarify can help detect bias in the incoming data, which is an important aspect of maintaining data quality and ensuring fair model performance.
- D. Incorrect.
Retraining the model without monitoring data drift may solve performance degradation temporarily, but it does not address the root cause, such as data quality issues.
- E. Correct.
Amazon SageMaker Model Monitor can track prediction drift, which is critical for understanding how the model's predictions deviate over time from expected behavior.