SHapley Additive exPlanations

A form of Model Agnostic Interpretability Based on Cooperative Game Theory Each feature has an importance value for prediction (Shapley Values)

Shapely Value for Feature

Shapely valyue for feature $i$ is:

$${\varphi }_i=\sum _{i=1}^n\left[S|\frac{!(M-|S|-1)!}{M}\right]\times [f(S\cup \{i\})-f(S)]$$

Variants

• TreeSHAP
• DeepSHAP
• LinearSHAP
• KernelSHAP
• PartitionSHAP