AWS Certified Machine Learning – SpecialtyModelingHard
A data scientist is developing a predictive model for customer churn. The business stakeholders emphasize the importance of understanding *why* a customer is predicted to churn, not just the prediction itself, to enable targeted intervention strategies. The model currently achieves high accuracy, but it's a complex ensemble model (e.g., Gradient Boosting) that is difficult to interpret. Which interpretability technique is most suitable for explaining individual predictions of this complex model to business users?
- AUse SHAP (SHapley Additive exPlanations) to provide global model explanations.
- BCreate global feature importance plots for the entire model.
- CSimplify the model by replacing the ensemble with a linear regression.
- DApply LIME (Local Interpretable Model-agnostic Explanations) to explain individual predictions.
Show answer & explanationAnswer & explanation
Correct answer: D. Apply LIME (Local Interpretable Model-agnostic Explanations) to explain individual predictions.
The requirement is to understand *why* an *individual customer* is predicted to churn. LIME is a model-agnostic technique specifically designed to explain individual predictions by creating a local, interpretable approximation around the prediction point, making it ideal for explaining complex models to business users on a case-by-case basis.
Why the other options are wrong
- A. While SHAP can provide local explanations, the question asks for the *most suitable* for business users and LIME is often cited for its intuitive local explanations, especially when compared to global SHAP explanations as an option. SHAP also provides individual explanations, but the option specifically mentions global explanations for SHAP, making it less suitable than LIME for *individual* prediction explanations in this context.
- B. Global feature importance shows overall influence, but not *why* a specific customer churns.
- C. Simplifying the model might reduce accuracy, and the goal is to explain the existing complex model, not replace it.
LIME (Local Interpretable Model-agnostic Explanations)
A technique that explains the predictions of any machine learning model by approximating it locally with an interpretable model.
- Model-agnostic: Works with any black-box model.
- Provides local explanations: Explains a single prediction.
- Generates an interpretable model (e.g., linear model) around the instance of interest.
Memory trick: LIME explains local predictions, like zooming in.