A model-agnostic explainability method that attributes predictions to input features using game-theoretic Shapley values computed via weighted local regression.