Databricks Machine Learning Professional Question 230
Select 2A company has deployed a machine learning model to predict customer churn. The model was trained on historical data where customer interactions were largely in-person. However, after the deployment, the company shifted to a primarily online interaction model due to business changes. Which type(s) of data drift is/are most likely to occur in this scenario?
- A
Feature drift, because the way customer behavior is captured has changed significantly.
- B
Label drift, because the definition of churn may have changed after the transition to online interactions.
- C
Feature drift, because changes in customer interaction channels could lead to shifts in the distribution of input features.
- D
Label drift, because the change in customer behavior does not directly influence the input features.
Show answer and explanation
Correct answers: A, C
Explanation
Feature drift happens when the distribution of input features changes over time, as is likely when the data collection process changes (e.g., in-person to online interactions). Label drift occurs when the distribution or definition of the target variable changes, which might happen if the criteria for identifying churn changes after a business process shift.
- A. Correct.
Feature drift is likely because the transition from in-person to online interactions changes how certain features (e.g., frequency of interactions, time spent, or methods of contact) are recorded or distributed.
- B. Incorrect.
Label drift can occur if the company redefines what constitutes customer churn (e.g., churn might be identified differently for online interactions compared to in-person interactions).
- C. Correct.
Feature drift is a valid possibility because a shift in interaction channels (e.g., in-person to online) can lead to changes in the input data distribution.
- D. Incorrect.
This is incorrect because label drift is not related to changes in input features but rather to changes in the distribution or definition of the target variable.