MLS-C01 Question 230
Select 2You are building a binary classification model to predict whether users will subscribe to a premium service based on their activity data. After training the model, you are evaluating its performance. The model has a high accuracy of 95%, but the business team informs you that false negatives (predicting a user will not subscribe when they actually would) have significant financial consequences. Which evaluation metric(s) should you prioritize to address this concern?
- A
Precision
- B
Recall
- C
F1 Score
- D
Area Under the Curve (AUC)-Receiver Operating Characteristics (ROC)
- E
Accuracy
Show answer and explanation
Correct answers: B, C
Explanation
In this scenario, the business impact of false negatives is significant, so evaluation metrics that address this concern should be prioritized. Recall directly measures the ability to minimize false negatives, and the F1 score provides a balance between precision and recall. These metrics are more relevant than accuracy, precision, or AUC-ROC, which do not directly address the specific issue of false negatives or the need to balance errors effectively.
- A. Incorrect.
Precision is useful when false positives are more concerning than false negatives, but this scenario prioritizes addressing false negatives, making precision less relevant.
- B. Correct.
Recall is critical in this scenario, as it measures the model's ability to identify true positives and avoids missing users who would subscribe (false negatives).
- C. Correct.
F1 Score is a harmonic mean of precision and recall, and it is particularly useful when there is a need to balance precision and recall, especially in cases where false negatives and false positives carry different costs.
- D. Incorrect.
AUC-ROC measures the overall performance of a binary classifier and is useful for comparing models but does not directly address the concern about false negatives.
- E. Incorrect.
Accuracy is not appropriate here because it does not distinguish between types of errors. A high accuracy can still occur even if the model has many false negatives, which is problematic in this scenario.