Explaining predictive models with mixed features using Shapley values and conditional inference trees

It is becoming increasingly important to explain complex, black-box machine\nlearning models. Although there is an expanding literature on this topic,\nShapley values stand out as a sound method to explain predictions from any type\nof machine learning model. The original development of Shapley values for\nprediction explanation relied on the assumption that the features being\ndescribed were independent. This methodology was then extended to explain\ndependent features with an underlying continuous distribution. In this paper,\nwe propose a method to explain mixed (i.e. continuous, discrete, ordinal, and\ncategorical) dependent features by modeling the dependence structure of the\nfeatures using conditional inference trees. We demonstrate our proposed method\nagainst the current industry standards in various simulation studies and find\nthat our method often outperforms the other approaches. Finally, we apply our\nmethod to a real financial data set used in the 2018 FICO Explainable Machine\nLearning Challenge and show how our explanations compare to the FICO challenge\nRecognition Award winning team.\n