Having examined local surrogate models with LIME, this chapter introduces SHapley Additive exPlanations (SHAP), a different approach to understanding model predictions based on principles from cooperative game theory.

You will learn about the theoretical basis of Shapley values and how the SHAP framework adapts them to assign an importance value for each feature's contribution to a specific prediction. We will discuss key properties that make SHAP values a consistent and accurate measure of feature importance.

The chapter covers different methods for computing SHAP values:

• KernelSHAP: A model-agnostic method applicable to any machine learning model.
• TreeSHAP: An efficient method specifically designed for tree-based models like decision trees, Random Forests, and Gradient Boosting Machines.

Furthermore, you will learn how to implement SHAP using its Python library and interpret common visualizations like force plots, summary plots, and dependence plots to gain insights into both individual predictions (local explanations) and overall model behavior (global explanations). Practical examples will guide you through generating and understanding these explanations.