Shapley Additive Explanations for Knowledge Discovery via Surrogate Models