Generalized SHAP: Generating multiple types of explanations in machine learning

Many important questions about a model cannot be answered just by explaining\nhow much each feature contributes to its output. To answer a broader set of\nquestions, we generalize a popular, mathematically well-grounded explanation\ntechnique, Shapley Additive Explanations (SHAP). Our new method - Generalized\nShapley Additive Explanations (G-SHAP) - produces many additional types of\nexplanations, including: 1) General classification explanations; Why is this\nsample more likely to belong to one class rather than another? 2) Intergroup\ndifferences; Why do our model's predictions differ between groups of\nobservations? 3) Model failure; Why does our model perform poorly on a given\nsample? We formally define these types of explanations and illustrate their\npractical use on real data.\n