BIROn - Birkbeck Institutional Research Online

    Joint Shapley Values: a measure of joint feature importance

    Harris, C. and Pymar, Richard and Rowat, Colin (2022) Joint Shapley Values: a measure of joint feature importance. International Conference on Learning Representations ,

    [img]
    Preview
    Text
    47310a.pdf - Author's Accepted Manuscript

    Download (323kB) | Preview
    Official URL: https://iclr.cc/

    Abstract

    The Shapley value is one of the most widely used measures of feature importance partly as it measures a feature's average effect on a model's prediction. We introduce joint Shapley values, which directly extend Shapley's axioms and intuitions: joint Shapley values measure a set of features' average effect on a model's prediction. We prove the uniqueness of joint Shapley values, for any order of explanation. Results for games show that joint Shapley values present different insights from existing interaction indices, which assess the effect of a feature within a set of features. The joint Shapley values seem to provide sensible results in ML attribution problems. With binary features, we present a presence-adjusted global value that is more consistent with local intuitions than the usual approach.

    Metadata

    Item Type: Article
    Additional Information: The Tenth International Conference on Learning Representations, ICLR 2022
    School: School of Business, Economics & Informatics > Economics, Mathematics and Statistics
    Depositing User: Richard Pymar
    Date Deposited: 18 Mar 2022 06:15
    Last Modified: 20 Mar 2022 05:45
    URI: https://eprints.bbk.ac.uk/id/eprint/47310

    Statistics

    Activity Overview
    6 month trend
    7Downloads
    6 month trend
    26Hits

    Additional statistics are available via IRStats2.

    Archive Staff Only (login required)

    Edit/View Item Edit/View Item