Databricks Machine Learning Associate Question 260
Single answerYou are training a regression model in Databricks, where the target variable has been log-transformed to handle skewness in its distribution. After evaluating the model, you notice that the RMSE (Root Mean Squared Error) value appears smaller than expected. What could be the most likely explanation for this observation?
- A
The RMSE has not been exponentiated back to the original scale of the target variable.
- B
The RMSE calculation is incorrect, and the model needs to be retrained.
- C
The model is overfitting to the training data, resulting in an artificially low RMSE.
- D
The log-transformation of the target variable inherently reduces the RMSE.
Show answer and explanation
Correct answer: A
Explanation
When the target variable is log-transformed, metrics like RMSE are computed in the log-transformed space. To interpret these metrics on the original scale, the log-transformed RMSE must be exponentiated. Failing to do so results in RMSE values that appear artificially small relative to the original scale of the target variable.
- A. Correct.
Correct. When the target variable is log-transformed, the RMSE is computed in the log-transformed space. To interpret RMSE on the original scale, it needs to be exponentiated. Otherwise, it will appear smaller than expected.
- B. Incorrect.
Incorrect. There is no indication that the RMSE calculation itself is incorrect; it simply has not been exponentiated back to the original scale.
- C. Incorrect.
Incorrect. While overfitting can lead to low RMSE on training data, this is not relevant to the issue of interpreting RMSE when the target variable is log-transformed.
- D. Incorrect.
Incorrect. Log-transforming the target variable does not inherently reduce RMSE. The smaller value is due to the RMSE being computed in the log-transformed space.