Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Preference-Based Abstract Argumentation for Case-Based Reasoning (with Appendix)

Adam Gould, Guilherme Paulino-Passos, Seema Dadhania, Matthew Williams, Francesca Toni

TL;DR

The paper addresses interpretable, data-driven classification by integrating user-defined preferences into Abstract Argumentation for Case-Based Reasoning (AA-CBR) through a new framework AA-CBR-$\mathcal{P}$. It formalizes a sequence of preorders $\mathcal{P}$ that govern how cases are compared and attacks are determined, ensuring predictions respect these preferences, and proves nearest- and preferred-case properties under coherent data. The approach is demonstrated on a BrainWear clinical dataset, where AA-CBR-$\mathcal{P}$ with targeted feature-preferences and stages outperforms interpretable baselines while offering richer explanations. This work advances explainable healthcare AI by enabling flexible, domain-specific preference incorporation into case-based, argumentation-based reasoning.

Abstract

In the pursuit of enhancing the efficacy and flexibility of interpretable, data-driven classification models, this work introduces a novel incorporation of user-defined preferences with Abstract Argumentation and Case-Based Reasoning (CBR). Specifically, we introduce Preference-Based Abstract Argumentation for Case-Based Reasoning (which we call AA-CBR-P), allowing users to define multiple approaches to compare cases with an ordering that specifies their preference over these comparison approaches. We prove that the model inherently follows these preferences when making predictions and show that previous abstract argumentation for case-based reasoning approaches are insufficient at expressing preferences over constituents of an argument. We then demonstrate how this can be applied to a real-world medical dataset sourced from a clinical trial evaluating differing assessment methods of patients with a primary brain tumour. We show empirically that our approach outperforms other interpretable machine learning models on this dataset.

Preference-Based Abstract Argumentation for Case-Based Reasoning (with Appendix)

TL;DR

The paper addresses interpretable, data-driven classification by integrating user-defined preferences into Abstract Argumentation for Case-Based Reasoning (AA-CBR) through a new framework AA-CBR-. It formalizes a sequence of preorders that govern how cases are compared and attacks are determined, ensuring predictions respect these preferences, and proves nearest- and preferred-case properties under coherent data. The approach is demonstrated on a BrainWear clinical dataset, where AA-CBR- with targeted feature-preferences and stages outperforms interpretable baselines while offering richer explanations. This work advances explainable healthcare AI by enabling flexible, domain-specific preference incorporation into case-based, argumentation-based reasoning.

Abstract

In the pursuit of enhancing the efficacy and flexibility of interpretable, data-driven classification models, this work introduces a novel incorporation of user-defined preferences with Abstract Argumentation and Case-Based Reasoning (CBR). Specifically, we introduce Preference-Based Abstract Argumentation for Case-Based Reasoning (which we call AA-CBR-P), allowing users to define multiple approaches to compare cases with an ordering that specifies their preference over these comparison approaches. We prove that the model inherently follows these preferences when making predictions and show that previous abstract argumentation for case-based reasoning approaches are insufficient at expressing preferences over constituents of an argument. We then demonstrate how this can be applied to a real-world medical dataset sourced from a clinical trial evaluating differing assessment methods of patients with a primary brain tumour. We show empirically that our approach outperforms other interpretable machine learning models on this dataset.
Paper Structure (27 sections, 5 theorems, 2 equations, 6 figures, 2 tables)

This paper contains 27 sections, 5 theorems, 2 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Abstract Argumentation
  4. Abstract Argumentation for Case-Based Reasoning (AA-CBR)
  5. AA-CBR with Stages
  6. Motivations
  7. Preference-Based Abstract Argumentation for Case-Based Reasoning
  8. Nearest and Preferred Cases
  9. Empirical Evaluation
  10. Methodology
  11. Evaluation
  12. Related Work and Discussion
  13. Conclusion
  14. Theorem Proofs
  15. Proof of Theorem \ref{['theorem:preferred-cases']}
  16. ...and 12 more sections

Key Result

Theorem 1

There exists a sequence of partial orders $\mathcal{P}$ and casebase $D$, such that there is no partial order, $\succcurlyeq$, in which $AF_{}(D, x_{N})$ constructed with $\succcurlyeq$ is the same as $AF_{\mathcal{P}}(D, x_{N})$ constructed with $\mathcal{P}$ and $\hbox{AA-CBR$(D, x_{N})$}$ is the

Figures (6)

  • Figure 1: A comparison of scenarios utilising AA-CBR with a naive attempt to introduce preferences
  • Figure 2: An AF derived from AA-CBR with Stages as defined in Definition \ref{['def:aa-cbr-with-stages']}. Despite $C'_1$ having features and stages equal to $N'$, the outcome predicted for $N'$ is $-$, which differs from $C'_1$'s outcome. The nearest cases property does not hold.
  • Figure 3: Two AFs showcasing potential attacks that are blocked by more concise cases according to the two concision conditions
  • Figure 4: A comparison of scenarios utilising AA-CBR-$\mathcal{P}$ to introduce preferences as defined in Example \ref{['example:aa-cbr-methodology-1']}.
  • Figure 5: $AF_{\mathcal{P}{\langle \supseteq, \sqsupseteq \rangle}}(D, x_{N})$ as defined in Example \ref{['example:properties-example-1']}. The nearest cases property holds. $C'_{1}$ is not prevented from attacking as $C'_{3}$ is no longer considered a more concise case.
  • ...and 1 more figures

Theorems & Definitions (27)

  • Definition 1: Adapted from monotonicity-and-noise-toleranceDEAr
  • Definition 2: Adapted from monotonicity-and-noise-toleranceDEAr
  • Definition 3: Adapted from arbitrated-argumentative-dispute
  • Example 1
  • Example 2
  • Example 3
  • Definition 4: Preference Ordering
  • Definition 5: Potential Attacks
  • Definition 6: Casebase Attacks
  • Definition 7: Coherent and Incoherent Casebases
  • ...and 17 more