SHAP (SHapley Additive exPlanations) is a game-theoretic approach that assigns each feature an importance score for a particular prediction — based on Shapley values from cooperative game theory, providing consistent, locally accurate, and fair attribution of feature contributions.

How Does SHAP Work?

• Shapley Value: The average marginal contribution of a feature across all possible feature combinations.
• Additivity: Feature contributions sum to the difference between the prediction and the average prediction.
• Global + Local: SHAP provides both per-prediction (local) and dataset-wide (global) explanations.
• Implementations: TreeSHAP (fast for tree models), KernelSHAP (model-agnostic), DeepSHAP (deep learning).

Why It Matters

• Feature Ranking: SHAP importance plots show which process parameters most influence yield/defect predictions.
• Interaction Detection: SHAP interaction values reveal synergistic effects between process variables.
• Debugging: Identifies when models rely on unexpected features — flagging potential data leakage or confounders.

SHAP is the fair scorecard for features — using game theory to assign each process variable its fair share of credit for every prediction.