I understand that SHAP values have a property called additivity that means that if you add the SHAP value of each explanatory variable of a particular example to the average prediction of the model on the dataset you obtain the prediction for that particular example. I know that SHAP values are LOCAL explanations, therefore they do not exactly apply to multiple examples, so I assume the answer is no, they are not additive across examples.

The reason I am interested in this question is that I want to add up the SHAP values of several misclassified examples, does anyone has any insight on why the output may or may not be useful to explain what is the model doing with those examples?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.