AI0-001 Implementing AI Solutions Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Replace the gradient-boosted tree with a logistic regression model and report the model coefficients as the explanation for every decision.

  Why it's wrong here

  A logistic regression model is inherently more transparent, but replacing the model sacrifices accuracy and still does not provide a per-applicant explanation that accounts for feature interactions. Global coefficients describe average behavior, not the specific contribution of each feature for a given application. Regulators typically require decision-specific reasoning, so swapping the model alone is insufficient and may not satisfy the audit requirement.

• ✗

  Provide the raw input features and the final score to the regulator and let them interpret the decision themselves.

  Why it's wrong here

  Supplying raw features and a score shifts the interpretive burden to the regulator and does not constitute an explanation of the model's reasoning. The firm is required to explain why the decision was made, not merely to disclose inputs and outputs. Without attribution or a decision rule, the regulator cannot determine which factors drove the outcome. This approach fails the explainability and auditability requirement.

• ✗

  Report the model's global feature importance ranking from the training run as the explanation for each decision.

  Why it's wrong here

  Global feature importance describes which features matter most across the entire training set, not why a specific application was approved or denied. Two applicants with different profiles can receive different decisions for reasons that are invisible in a global ranking. Regulators examining an individual case need decision-specific reasoning, so global importance alone does not meet the auditability requirement and could be misleading in a review.

• ✓

  Use SHAP values computed for each individual application to attribute the model's output to its input features, and present those attributions as the explanation.

  Why this is correct

  SHAP values provide a per-instance, additive attribution of the model output to each input feature, which is exactly what is needed to explain an individual decision. They work with gradient-boosted trees and other complex models, and they are grounded in cooperative game theory, giving a consistent and locally accurate explanation. This satisfies the auditability requirement without sacrificing model accuracy, and the attributions can be logged and reviewed.