What is a Surrogate Model?
Surrogate model is an explainability technique where you build a transparent model off of the predictions of an actual model. The model is built in parallel with the model data. It is used when an outcome of interest cannot be easily directly measured, so a model of the outcome is used instead.
Teams often find the surrogate model uuseful if they don’t have the original model to extract SHAP; instead, you can build a surrogate off of model decisions. It’s less ideal for regulatory use-cases as it is highly dependent on the data the model sees.