MLA-C01 Question 202
Select 3You have trained a binary classification model using Amazon SageMaker. After evaluating the model's performance, you observe the following metrics on the validation dataset: Accuracy = 0.92, Precision = 0.95, Recall = 0.60. Which of the following actions would help improve the model's performance for a use case where minimizing false negatives is critical?
- A
Increase the threshold for classifying a positive prediction.
- B
Decrease the threshold for classifying a positive prediction.
- C
Collect and include more samples of the minority class in the training dataset.
- D
Use a different evaluation metric, such as F1-score, to guide model improvements.
- E
Focus on improving the precision further, as it is currently lower than the accuracy.
Show answer and explanation
Correct answers: B, C, D
Explanation
In scenarios where minimizing false negatives is critical, recall is the most important metric to focus on. Decreasing the threshold for positive predictions increases recall, as more positive cases are identified. Incorporating more samples of the minority class helps the model better understand and predict positive cases. Additionally, using a balanced metric like F1-score ensures that recall is adequately prioritized while maintaining reasonable precision.
- A. Incorrect.
Increasing the threshold for classifying a positive prediction would likely decrease recall further, as fewer positive predictions would be made, which is counterproductive when minimizing false negatives is critical.
- B. Correct.
Decreasing the threshold for classifying a positive prediction would increase recall by capturing more true positives, helping to minimize false negatives.
- C. Correct.
Collecting and including more samples of the minority class would help the model learn to better identify positive cases, thereby improving recall and reducing false negatives.
- D. Correct.
Using a different evaluation metric, such as F1-score, which balances precision and recall, could help guide model improvements to ensure that recall is not overlooked.
- E. Incorrect.
Focusing on improving precision further would not directly address the issue of high false negatives, as precision is already high and recall is the more relevant metric for this scenario.