SHAP Values & Feature Attribution - Interactive Visualization

SHAP values are rooted in cooperative game theory: each feature's contribution is the average marginal effect of including that feature across all possible orderings of features. The waterfall plot shows exactly how each feature pushed a prediction above or below the population average. Unlike traditional feature importance, SHAP gives directional, per-prediction explanations.

• Read a waterfall plot: starting from the base value, each bar shows a feature pushing the prediction up (red) or down (blue)
• Understand why SHAP values satisfy three desirable properties: local accuracy, missingness, and consistency
• Compare SHAP feature importance (mean absolute SHAP) vs permutation importance - they often disagree on borderline features
• See how the same feature can have positive SHAP for one data point and negative for another
• Learn TreeSHAP: the fast O(TLD) algorithm that makes SHAP practical for gradient boosting models

Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.