This paper introduces SHAP (SHapley Additive exPlanations), a unified framework for interpreting model predictions. SHAP assigns each feature an importance value for a particular prediction, unifying six existing methods and providing a unique solution with desirable properties. The framework is based on game theory and defines additive feature attribution methods as linear functions of binary variables. SHAP values are derived from the Shapley values of a conditional expectation function of the original model, ensuring local accuracy, missingness, and consistency. The paper presents new methods for estimating SHAP values, including Kernel SHAP and Deep SHAP, which improve computational performance and align better with human intuition. Experiments show that SHAP values are more consistent with human intuition than other methods and provide better explanations of model outputs. The paper also discusses the implications of SHAP for model interpretability and the development of future methods.This paper introduces SHAP (SHapley Additive exPlanations), a unified framework for interpreting model predictions. SHAP assigns each feature an importance value for a particular prediction, unifying six existing methods and providing a unique solution with desirable properties. The framework is based on game theory and defines additive feature attribution methods as linear functions of binary variables. SHAP values are derived from the Shapley values of a conditional expectation function of the original model, ensuring local accuracy, missingness, and consistency. The paper presents new methods for estimating SHAP values, including Kernel SHAP and Deep SHAP, which improve computational performance and align better with human intuition. Experiments show that SHAP values are more consistent with human intuition than other methods and provide better explanations of model outputs. The paper also discusses the implications of SHAP for model interpretability and the development of future methods.