SHAP Explainer
Fair feature attribution via Shapley values — local explanations and global importance in one tool.
SHAP (SHapley Additive exPlanations) borrows a concept from cooperative game theory: if features are 'players' and the prediction is the 'payout', Shapley values give each feature a fair share. The sum of all SHAP values plus the base value always equals the prediction — no credit is lost or invented.
In the Local tab, drag the feature sliders to set an instance, then hit 'Compute SHAP'. The force plot shows how each feature pushes the prediction away from the average. The waterfall breaks it down feature by feature. Click any feature to see every coalition and its marginal contribution — the raw data behind the Shapley formula.
Switch to the Global tab to see which features matter most on average (mean |SHAP|) and how each feature's value relates to its effect (beeswarm). Blue dots are low feature values, red dots are high — the horizontal spread tells you how much that feature can move the prediction.