SHAP Values Explainer
SHAP values are feature attributions based on Shapley values from cooperative game theory. They treat features as participants and distribute the difference between a model’s prediction and a reference value among those features.
Conceptually, the method examines how a feature changes the prediction across many possible groups of other features. Exact calculation is often expensive, so practical tools use model-specific algorithms, sampling, or approximations.
SHAP can provide local explanations for individual predictions and aggregate patterns across a dataset. The meaning depends on how missing features are represented and which background distribution defines the reference.
Correlated features can make allocation difficult to interpret, and attributions do not establish causation. Reports should document the explainer type, background data, output scale, approximation settings, and stability across plausible choices.
SHAP values distribute the difference between a model's prediction and a reference value among input features using Shapley-value principles.
Key terms include Shapley value, coalition, feature contribution, reference value, background distribution, local explanation, approximation, correlation and output scale.
Each feature receives credit based on its marginal contribution across many coalitions.
The contributions add to the gap between the prediction and its reference.
A different reference population can produce different SHAP values.
Correlated features and approximation choices should be disclosed.
Explore more "Explainers"
Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.