Shapley Sets: Feature Attribution via Recursive Function Decomposition

TL;DR

Shapley Sets introduces a novel feature attribution method that assigns value to groups of features, effectively handling complex feature interactions and dependencies, and offers a fairer and more accurate explanation than traditional Shapley values.

Contribution

It proposes Shapley Sets, a new recursive decomposition algorithm that attributes value to feature groups, addressing limitations of traditional Shapley value methods in complex data.

Findings

Shapley Sets decomposes models with log-linear complexity.

It is equivalent to Shapley value over transformed features.

Shapley Sets improves attribution accuracy in complex dependency scenarios.

Abstract

Despite their ubiquitous use, Shapley value feature attributions can be misleading due to feature interaction in both model and data. We propose an alternative attribution approach, Shapley Sets, which awards value to sets of features. Shapley Sets decomposes the underlying model into non-separable variable groups using a recursive function decomposition algorithm with log linear complexity in the number of variables. Shapley Sets attributes to each non-separable variable group their combined value for a particular prediction. We show that Shapley Sets is equivalent to the Shapley value over the transformed feature set and thus benefits from the same axioms of fairness. Shapley Sets is value function agnostic and we show theoretically and experimentally how Shapley Sets avoids pitfalls associated with Shapley value based alternatives and are particularly advantageous for data types with complex dependency structure.