Applications of cooperative game theory to interpretable machine learning
Abstract/Contents
- Abstract
- Model-agnostic feature importance measures are central to the task of demystifying opaque or "black-box" machine learning models. The proliferation of such models within high-stakes decision making settings such as healthcare or banking necessitates the development of flexible and trustworthy approaches to the problem. With no ground truth feature importance to compare to, competing methods provide contrasting approaches and/or philosophies often with a claim of superiority. Some of the most popular recent approaches are adaptations of tools from cooperative game theory used in reward or cost sharing problems. In this document, we report on recent advances among such feature importance methods. In particular, we discuss a "data-centric" cohort-based framework for model-agnostic local feature importance using Shapley values. We propose a primary importance measure and explore several adaptations of that method better suited for specific use cases or data regimes. We analyze the properties and behaviors of these methods and apply them to a broad range of synthetic and real-world problem settings including voter registration and recidivism data. We then propose and discuss new methods for local importance aggregation and feature importance evaluation.
Description
Type of resource | text |
---|---|
Form | electronic resource; remote; computer; online resource |
Extent | 1 online resource. |
Place | California |
Place | [Stanford, California] |
Publisher | [Stanford University] |
Copyright date | 2023; ©2023 |
Publication date | 2023; 2023 |
Issuance | monographic |
Language | English |
Creators/Contributors
Author | Seiler, Benjamin Bradbury |
---|---|
Degree supervisor | Owen, Art B |
Thesis advisor | Owen, Art B |
Thesis advisor | Palacios Roman, Julia Adela |
Thesis advisor | Taylor, Jonathan E |
Degree committee member | Palacios Roman, Julia Adela |
Degree committee member | Taylor, Jonathan E |
Associated with | Stanford University, School of Humanities and Sciences |
Associated with | Stanford University, Department of Statistics |
Subjects
Genre | Theses |
---|---|
Genre | Text |
Bibliographic information
Statement of responsibility | Benjamin B. Seiler. |
---|---|
Note | Submitted to the Department of Statistics. |
Thesis | Thesis Ph.D. Stanford University 2023. |
Location | https://purl.stanford.edu/xq291xs8637 |
Access conditions
- Copyright
- © 2023 by Benjamin Bradbury Seiler
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...