Google Professional Machine Learning Engineer Question 504
Select 2Google Cloud PlatformYou have deployed a machine learning model to predict customer churn based on input features such as account age, monthly usage, and support ticket interactions. You want to monitor for feature attribution drift to ensure that the importance of individual features as perceived by the model does not change significantly over time. Which of the following steps should you take to effectively monitor feature attribution drift?
- A
Use SHAP values to compute feature importance scores and compare them over time.
- B
Set up a monitoring pipeline to trigger alerts when feature distributions deviate from training data.
- C
Periodically calculate and compare the distribution of model predictions across different time windows.
- D
Integrate Explainable AI tools in Google Cloud to analyze feature importance trends.
- E
Monitor the mean absolute error (MAE) of the model predictions over time.
Show answer and explanation
Correct answers: A, D
Explanation
Feature attribution drift occurs when the importance of features as perceived by the model changes over time. This can result in a misalignment between the model's behavior and the expected feature importance. Using SHAP values or Explainable AI tools in Google Cloud allows you to monitor and track feature importance trends over time, making these the correct approaches for monitoring feature attribution drift.
- A. Correct.
Using SHAP values is a standard approach to compute feature importance and analyze how it changes over time. This directly addresses feature attribution drift.
- B. Incorrect.
Monitoring feature distributions is important for feature drift but does not directly address feature attribution drift, which focuses on the relationship between features and model predictions.
- C. Incorrect.
Comparing distributions of model predictions helps detect prediction drift but does not provide insights into feature attribution drift.
- D. Correct.
Google Cloud's Explainable AI tools allow you to analyze feature importance trends, making it a relevant step for monitoring feature attribution drift.
- E. Incorrect.
Monitoring metrics like mean absolute error (MAE) can help detect performance degradation but does not directly relate to feature attribution drift.