MLS-C01 Question 149
Single answerYou are a Machine Learning Specialist working for a healthcare company. Your team has developed a black-box model to predict the likelihood of a patient being readmitted to the hospital within 30 days of discharge. However, the clinicians are hesitant to use the model without understanding how it makes predictions. Which approach can you use to improve the model's interpretability and build trust while maintaining predictive performance?
- A
Use SHAP (SHapley Additive exPlanations) values to explain the contribution of each feature to individual predictions.
- B
Replace the black-box model with a simple linear regression model, as linear models are inherently interpretable.
- C
Train an interpretable surrogate model (e.g., decision tree) on the outputs of the black-box model to approximate its behavior.
- D
Enable AWS CloudTrail to log API calls made by the black-box model for better auditing and interpretability.
Show answer and explanation
Correct answer: A
Explanation
When working with black-box models, explaining predictions is key to gaining stakeholder trust, especially in sensitive domains like healthcare. SHAP values are an industry-standard method for explaining individual predictions by attributing contributions to features while preserving the original model's predictive accuracy. This makes them highly suitable for balancing interpretability and performance.
- A. Correct.
SHAP values are specifically designed to provide explanations for black-box models by quantifying the contribution of each feature to the model's predictions, making this the best approach to improve model interpretability while maintaining predictive performance.
- B. Incorrect.
Replacing the black-box model with a linear regression model sacrifices predictive performance for interpretability, which may not be acceptable in a healthcare context where accuracy is critical.
- C. Incorrect.
Training a surrogate model can approximate the behavior of the black-box model but does not provide precise explanations for individual predictions. This approach might lead to oversimplified or misleading interpretations.
- D. Incorrect.
AWS CloudTrail is used for logging and auditing API calls but does not provide any insights into the internal workings or predictions of a machine learning model.