A lender must tell each rejected applicant the main factors behind that specific decision. The model is a gradient boosted classifier deployed on Vertex AI. Which technique provides this?
Show answer and explanation
Answer: B
Per-prediction attributions show how each input pushed a particular score up or down, which supports individual adverse-action explanations. Vertex Explainable AI returns these attributions for deployed models. Global importance describes the model overall, a raw probability does not say which features drove it, and averaged SHAP values are global summaries rather than explanations of one decision.
Why the other options are wrong
- A. Global importance from validation data describes the model in general. It does not explain why this applicant received this outcome.
- C. A probability score says how confident the model is, but not which features drove the decision. Regulators expect specific reasons.
- D. Averaging across decisions produces global patterns. It does not isolate the factors that affected one applicant.