Google Professional Machine Learning Engineer Question 417
Single answerGoogle Cloud PlatformYou manage a machine learning model for predicting customer churn on Google Cloud. The model’s predictions are used to trigger customer retention strategies. You notice that the model's accuracy has been decreasing over time due to changes in customer behavior. Which retraining policy would be most appropriate in this scenario?
- A
Retrain the model on a fixed schedule, such as every month, regardless of performance.
- B
Retrain the model whenever there is a significant drop in key performance metrics, such as accuracy or precision.
- C
Retrain the model only when new features are added to the dataset.
- D
Retrain the model periodically but also monitor for data drift to determine if additional retraining is needed.
Show answer and explanation
Correct answer: B
Explanation
The most appropriate retraining policy in this scenario is to retrain the model whenever there is a significant drop in key performance metrics. This ensures that the model adapts to changes in customer behavior and maintains its predictive accuracy, which is critical for triggering effective customer retention strategies. Fixed schedules or retraining only with new features do not adequately address the specific issue of performance decline.
- A. Incorrect.
Retraining on a fixed schedule can be wasteful and may not address the performance issues caused by changes in customer behavior. This approach does not account for the actual performance of the model.
- B. Correct.
Retraining based on a significant drop in key performance metrics is a more dynamic approach and directly addresses the issue of decreasing accuracy due to changing customer behavior.
- C. Incorrect.
Retraining only when new features are added ignores the problem of model performance decay caused by shifts in data distribution or customer behavior. It is not suitable for this scenario.
- D. Incorrect.
While monitoring for data drift is a good practice, relying solely on a periodic retraining policy is not sufficient to address the performance drop caused by customer behavior changes.