MLS-C01 Question 231
Select 2You are building a binary classification model to detect fraudulent transactions. After evaluating the model, you obtain the following confusion matrix:
| Predicted: Fraudulent | Predicted: Not Fraudulent | |
|---|---|---|
| Actual: Fraudulent | 85 | 15 |
| Actual: Not Fraudulent | 20 | 380 |
Which metric or metrics can be accurately interpreted as a strength of the model's ability to correctly identify legitimate (non-fraudulent) transactions?
- A
Precision
- B
Recall
- C
Specificity
- D
Accuracy
- E
F1 Score
Show answer and explanation
Correct answers: C, D
Explanation
Specificity and accuracy are the most appropriate metrics for evaluating a model's ability to correctly identify legitimate transactions. Specificity directly measures the proportion of actual non-fraudulent transactions that are correctly classified as 'Not Fraudulent'. Accuracy provides an overall measure of correct classifications, which includes legitimate transactions. While metrics like precision and recall are valuable for assessing fraud detection, they do not specifically measure performance on legitimate transactions.
- A. Incorrect.
Precision measures how many predicted fraudulent transactions are actually fraudulent. It does not directly measure the model's ability to identify legitimate transactions.
- B. Incorrect.
Recall measures the model's ability to identify actual fraudulent transactions. It does not directly relate to the identification of legitimate transactions.
- C. Correct.
Specificity measures the ability of the model to correctly identify non-fraudulent (legitimate) transactions as 'Not Fraudulent', making it a key metric for this scenario.
- D. Correct.
Accuracy measures the overall proportion of correctly classified transactions (both fraudulent and non-fraudulent). Since legitimate transactions dominate in this dataset, accuracy is also a relevant measure of performance.
- E. Incorrect.
F1 Score is the harmonic mean of precision and recall, focusing on the balance between identifying fraudulent transactions correctly and minimizing false positives. However, it does not directly evaluate the model's performance in identifying legitimate transactions.