The GSHAP module is a direct extension to Scott Lundberg and Su-In Lee's kernel Shap.

The GSHAP module computes global SHAP values, based on the approach used by Scott Lundberg and Su-In Lee in their paper "A Unified Approach to Interpreting Model Predictions".

The GSHAP method uses the kernel SHAP method, in order to compute global SHAP values. The benefit herein is, the computation of global SHAP values in common and thereby comparable units. Thus there is no need of scaling the features, to make global feature contribution measures comparable. These are in the unit of a score measure specified by the user. The choices are:

• accuracy score
• recall score
• precision score
• area under the roc score

• model selection, in order to decide on a model with the most coherent explanation as suggested by domain expertise
• feature selection, with respect to a certain score measure that is of most importance given the use case