Paper: arXiv 2408.06679

Authors: Gregory Yampolsky, Dhruv Desai, Mingshu Li, Stefano Pasquali, Dhagash Mehta

Abstract

The explainability of black-box machine learning algorithms, commonly known as Explainable Artificial Intelligence (XAI), has become crucial for financial and other regulated industrial applications due to regulatory requirements and the need for transparency in business practices. Among the various paradigms of XAI, Explainable Case-Based Reasoning (XCBR) stands out as a pragmatic approach that elucidates the output of a model by referencing actual examples from the data used to train or test the model. Despite its potential, XCBR has been relatively underexplored for many algorithms such as tree-based models until recently. We start by observing that most XCBR methods are defined based on the distance metric learned by the algorithm. By utilizing a recently proposed technique to extract the distance metric learned by Random Forests (RFs), which is both geometry- and accuracy-preserving, we investigate various XCBR methods. These methods amount to identify special points from the training datasets, such as prototypes, critics, counter-factuals, and semi-factuals, to explain the predictions for a given query of the RF. We evaluate these special points using various evaluation metrics to assess their explanatory power and effectiveness.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 6.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper introduces a novel, mathematically dense methodology based on geometric distance metrics (RF-GAP) and defines specialized loss functions for XCBR, while also testing the framework on both public and proprietary financial data with specific evaluation metrics.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Explain Random Forests using XCBR"] --> B["Core Methodology<br>Extract RF Distance Metric"]
  B --> C["Identify Explanatory Points<br>Prototypes, Critics, Counter-factuals, Semi-factuals"]
  C --> D["Evaluation & Outcomes<br>Assess Explanatory Power & Effectiveness"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes integrating Random Forest proximities into the Explainable Case-Based Reasoning (XCBR) framework to identify prototypes, critics, semi-factuals, and counter-factuals for explaining RF predictions.

Source sentence
In the pursuit of refining the capabilities of XCBR for the predictions of RF model, we propose integrating RF proximities into the XCBR framework.

Assumes 2

  • The extracted Random Forest distance metric is geometry- and accuracy-preserving. model
    Source sentence
    By utilizing a recently proposed technique to extract the distance metric learned by Random Forests (RFs), which is both geometry- and accuracy-preserving
  • The datasets used (toy and funds) are representative for evaluating XCBR methods. data
    Source sentence
    In order to make sure the results are not biased towards one specific type of data, we compiled results from a variety of datasets.

Data

  • Diabetes · public other public
    Source sentence
    Diabetes: A small tabular dataset of 393 rows and two classes.
  • Blood alcohol · public other public
    Source sentence
    Blood alcohol: Larger tabular dataset of 2002 rows. A mixture of numeric and categorical columns.
  • German credit · public other public
    Source sentence
    German credit: A dataset with many rows that are a combination of numeric and categorical.
  • MNIST 4&9 · public other public
    Source sentence
    MNIST 4&9: MNIST dataset from sklearn filtered only for labels 4 and 9.
  • Funds Data · Morningstar fundamentals commercial
    march 2024 snapshot · 4002 funds
    Source sentence
    We sourced our data for funds from Morningstar Data warehouse data feed

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • GAP proximity delivers high-quality results in identifying and distinguishing prototypes and critics across datasets, with prototypes showing lower diversity and critics higher diversity. statistical
    Source sentence
    These findings lead to the conclusion that GAP proximity delivers high-quality results in identifying and distinguishing prototypes and critics across different various, demonstrating its effectiveness and accuracy.
  • Semi-factuals generally exhibit higher distance values compared to counter-factuals, confirming expected behavior. statistical
    Source sentence
    Semi-factuals generally exhibit higher distance values compared to counter-factuals. This aligns with the expectation that semi-factuals are further from the query point
  • In the MNIST visualization, the optimal number of prototypes is 2 with an f1 score of 0.997, indicating strong representation of the classes. performance
    Source sentence
    The f1 score is high, 0.997, indicating that the prototypes effectively represent the data
  • Evaluation metrics for funds data show prototypes have lower outlier score (4944 vs 373 for critics) and lower diversity (0.667 vs 4.537 for critics). statistical
    Source sentence
    ... outlier score, diversity, and out-of-distribution (OOD) distance. Specifically, prototypes, which exhibit lower values in these metrics
  • The proposed method successfully identifies prototypes, critics, semi- and counter-factuals, demonstrating the utility of GAP proximity for explaining RF predictions. theoretical
    Source sentence
    various types of explanantia including prototypes, critics, semi- and counter-factuals, which play critical roles in explaining RF outcomes.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 17 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 2 papers in the archive

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.