Databricks Machine Learning Professional Question 227
Single answerWhile monitoring a machine learning model deployed in production, you notice that the accuracy of the model has been steadily decreasing over time. Upon further investigation, you find that the distribution of the target variable has changed significantly compared to the training dataset, while the input features have largely maintained their original distribution. Which of the following best explains the issue?
- A
Label drift is occurring, as the statistical distribution of the target variable has changed.
- B
Feature drift is occurring, as the statistical distribution of the input features has changed.
- C
Both label drift and feature drift are occurring simultaneously.
- D
The issue is unrelated to label or feature drift and likely caused by data preprocessing errors.
Show answer and explanation
Correct answer: A
Explanation
This scenario highlights label drift, which occurs when the statistical distribution of the target variable (labels) changes over time. This is distinct from feature drift, which refers to changes in the distribution of input features. Identifying whether the issue is caused by label drift, feature drift, or other factors is crucial for diagnosing and addressing performance drop in production models.
- A. Correct.
Correct. Label drift refers to a change in the statistical distribution of the target variable (or labels) over time, which matches the scenario described.
- B. Incorrect.
Incorrect. Feature drift occurs when the statistical distributions of input features change, but in this case, the input features have largely maintained their original distribution.
- C. Incorrect.
Incorrect. The scenario describes only a change in the target variable's distribution, not simultaneous changes in both the features and the target variable.
- D. Incorrect.
Incorrect. While data preprocessing errors can also impact model performance, the scenario explicitly states that the issue is related to the target variable's distribution changing over time.