Examples

1. 1.

   SHAP (SHapley Additive exPlanations, Lundberg and Lee 2017) is used by Experian to explain credit-score decisions, identifying which financial features contributed most to a lending decision - required by ECOA regulations for adverse-action notices.

2. 2.

   LIME (Local Interpretable Model-Agnostic Explanations) is integrated into IBM Watson OpenScale to provide per-prediction explanations for production ML models, flagging cases where the model relied on protected attributes.

3. 3.

   Google's LIT (Language Interpretability Tool) is used by NLP researchers to analyse attention patterns, saliency maps, and counterfactual inputs for BERT and T5 models, helping identify spurious correlations in training data.