Question 35
Domain 3: Model DevelopmentA regression model was trained using a log-transformed target variable. The model outputs predicted values on the log scale, and the analyst wants to report prediction errors in the target variable's original units. What should the analyst do before calculating those evaluation metrics?
Correct answer: B
Explanation
When a target variable is modeled on the log scale, predictions must be transformed back with exponentiation before calculating metrics or interpreting results in the original units. — official.txt
Why each option is right or wrong
A. Calculate the metrics directly on the log-scale predictions and report them as original-unit errors
Metrics on log-scale values are not in the target variable's original units.
B. Exponentiate the log-scale predictions so the metrics are computed in the target variable's original units
The model outputs predictions on the log scale, but the analyst wants errors in the target variable's original units. The source material states that log-transformed target or prediction values must be exponentiated before metric calculation or interpretation in original units, so the predictions must be exponentiated first.
C. Standardize the log-scale predictions before computing metrics in the target variable's original units
Standardizing does not convert log-scale values back to the original scale.
D. Average the log-scale predictions across observations before computing metrics in the target variable's original units
Averaging predictions does not change them from the log scale to the original scale.