ShapDBM: Exploring Decision Boundary Maps in Shapley Space

Forum topic · 小凯 · 2026-03-25

Summary

This paper, titled ShapDBM: Exploring Decision Boundary Maps in Shapley Space, by Luke Watkin, Daniel Archambault, and Alex Telea (arXiv:2603.22235, published March 2026), introduces a new technique for visualizing machine learning classification boundaries. Decision Boundary Maps (DBMs) are an effective tool for visualizing ML classification boundaries, but their quality depends heavily on the dimensionality reduction (DR) method and the high-dimensional space used. For complex ML datasets, dimensionality reduction can produce many mixed classes, resulting in DBMs that are hard to use. The proposed approach transforms the data space into Shapley space and computes dimensionality reduction on it before generating the DBM. Compared to standard DBMs computed directly from the original data, the resulting maps achieve similar or higher quality metric values and feature visibly more compact decision zones that are easier to explore. The paper is relevant to researchers in machine learning interpretability, scientific visualization, and dimensionality reduction.

Paper Overview

Field: Machine Learning
Authors: Luke Watkin, Daniel Archambault, Alex Telea
Published: 2026-03-23
arXiv: 2603.22235

Abstract

Decision Boundary Maps (DBMs) are an effective tool for visualising machine learning classification boundaries. Yet, DBM quality strongly depends on the dimensionality reduction (DR) technique and high dimensional space used for the data points. For complex ML datasets, DR can create many mixed classes which, in turn, yield DBMs that are hard to use.

The authors propose a new technique to compute DBMs by transforming data space into Shapley space and computing dimensionality reduction on it. Compared to standard DBMs computed directly from data, the resulting maps have similar or higher quality metric values, along with visibly more compact and easier-to-explore decision zones.

Key Points

Standard DBMs are sensitive to the choice of dimensionality reduction technique and the high-dimensional representation of the data.
Complex ML datasets often produce many mixed classes under DR, leading to cluttered, hard-to-use decision boundary maps.
ShapDBM instead performs DR in Shapley space, transforming the data before projection.
The resulting maps show similar or higher quality metric values versus standard DBMs, with more compact decision regions that are easier to explore.

---

*Source: arXiv:2603.22235*