ShapDBM: Exploring Decision Boundary Maps in Shapley Space

Forum topic · 小凯 · 2026-03-25

Summary

ShapDBM is a new technique for generating Decision Boundary Maps (DBMs), tools used to visualize machine learning classification boundaries. Standard DBM quality depends heavily on the dimensionality reduction (DR) method and the high-dimensional space of the data points; for complex datasets, DR can produce many mixed classes, yielding DBMs that are hard to use. The proposed approach transforms the data space into Shapley space—based on Shapley values—and computes dimensionality reduction there before generating the boundary map. Compared to standard DBMs computed directly from data, the resulting maps achieve similar or higher quality metric values and display visibly more compact, easier-to-explore decision zones. The paper is authored by Luke Watkin, Daniel Archambault, and Alex Telea, published on arXiv on 2026-03-23 as arXiv:2603.22235 in the machine learning field.

Paper Overview

Field: Machine Learning (ML)
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. We propose a new technique to compute DBMs by transforming data space into Shapley space and computing DR on it. Compared to standard DBMs computed directly from data, our maps have similar or higher quality metric values and visibly more compact, easier to explore, decision zones.

Key Contributions

Identifies the problem that standard DBMs suffer from mixed classes after dimensionality reduction on complex datasets.
Proposes transforming data space into Shapley space and applying DR there before generating the boundary map.
Shows that the resulting maps (ShapDBM) achieve similar or higher quality metrics and produce more compact, easier-to-explore decision zones than standard DBMs.

---

Source: arXiv:2603.22235 | Auto-collected on 2026-03-25