KomFi. Academy

medium · FRM Part 2 Current Issues

When using LIME (Local Interpretable Model-agnostic Explanations) to explain a specific rejection by a deep-learning underwriting engine, a validator notes that the local linear surrogate model has a low R^2 in the neighborhood of the instance.

What is the primary risk associated with this finding?

  1. The model will be unable to produce any SHAP values for that instance due to the lack of local convergence.
  2. The bank is in violation of the Basel III Pillar 1 requirements for model-agnostic explainability.
  3. The local explanation is 'unfaithful' to the complex model and may provide misleading reason codes to the applicant.
  4. The underlying deep-learning model has likely overfitted the training data and should be retrained with more regularization.

Sign up free to see the explanation and track your rank →

More FRM Part 2 Current Issues practice

All Current Issues questions · FRM Part 2 study guide · FRM Part 2 glossary · FRM Part 2 flashcards