SHAP
SHAP: An attribution method estimating each feature’s contribution using Shapley-value principles. It compares predictions across feature coalitions to distribute contribution relative to a reference. Approximations and feature dependencies affect interpretation. Examples include Local credit-score explanation, Aggregated global importance, KernelSHAP.
In italiano: SHAPLegge V · Overfitting, bias e collasso dei modelliIt compares predictions across feature coalitions to distribute contribution relative to a reference. Approximations and feature dependencies affect interpretation.
Examples
- Local credit-score explanation
- Aggregated global importance
- KernelSHAP