The Susceptibility of Example-Based Explainability Methods to Class Outliers

Ikhtiyor Nematov; Dimitris Sacharidis; Katja Hose; Tomer Sagi

doi:10.1007/978-3-032-25305-7_2

The Susceptibility of Example-Based Explainability Methods to Class Outliers

Ikhtiyor Nematov
, Dimitris Sacharidis
, Katja Hose
, Tomer Sagi

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

This study explores the impact of class outliers on the effectiveness of example-based explainability methods for black-box machine learning models. We reformulate existing explainability evaluation metrics, such as correctness and relevance, specifically for example-based methods, and introduce a new metric, distinguishability. Using these metrics, we highlight the shortcomings of current example-based explainability methods, including those who attempt to suppress class outliers. We conduct experiments on two datasets, a text classification dataset and an image classification dataset, and evaluate the performance of four state-of-the-art explainability methods. Our findings underscore the need for robust techniques to tackle the challenges posed by class outliers.

Original language	English
Title of host publication	Machine Learning and Principles and Practice of Knowledge Discovery in Databases - International Workshops of ECML PKDD 2024, Revised Selected Papers
Editors	Mattia Cerrato, Danguole Kalinauskaite, Mantas Lukoševicius, Kristina Šutiene, Mykola Pechenizkiy
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	19-26
Number of pages	8
ISBN (Print)	9783032253040
DOIs	https://doi.org/10.1007/978-3-032-25305-7_2
State	Published - 2026
Externally published	Yes
Event	24th Joint European Conference on Machine Learning and Knowledge Discovery in Databases, ECML PKDD 2024 - Vilnius, Lithuania Duration: 9 Sep 2024 → 13 Sep 2024

Publication series

Name	Communications in Computer and Information Science
Volume	2559 CCIS
ISSN (Print)	1865-0929
ISSN (Electronic)	1865-0937

Conference

Conference	24th Joint European Conference on Machine Learning and Knowledge Discovery in Databases, ECML PKDD 2024
Country/Territory	Lithuania
City	Vilnius
Period	9/09/24 → 13/09/24

Bibliographical note

Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

Keywords

explainability
explainability evaluation
interpretability

ASJC Scopus subject areas

General Computer Science
General Mathematics

Access to Document

10.1007/978-3-032-25305-7_2

Cite this

Nematov, I., Sacharidis, D., Hose, K., & Sagi, T. (2026). The Susceptibility of Example-Based Explainability Methods to Class Outliers. In M. Cerrato, D. Kalinauskaite, M. Lukoševicius, K. Šutiene, & M. Pechenizkiy (Eds.), Machine Learning and Principles and Practice of Knowledge Discovery in Databases - International Workshops of ECML PKDD 2024, Revised Selected Papers (pp. 19-26). (Communications in Computer and Information Science; Vol. 2559 CCIS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-032-25305-7_2

Nematov, Ikhtiyor ; Sacharidis, Dimitris ; Hose, Katja et al. / The Susceptibility of Example-Based Explainability Methods to Class Outliers. Machine Learning and Principles and Practice of Knowledge Discovery in Databases - International Workshops of ECML PKDD 2024, Revised Selected Papers. editor / Mattia Cerrato ; Danguole Kalinauskaite ; Mantas Lukoševicius ; Kristina Šutiene ; Mykola Pechenizkiy. Springer Science and Business Media Deutschland GmbH, 2026. pp. 19-26 (Communications in Computer and Information Science).

@inproceedings{c4797001ef7440df84fdc4af2cffb2b4,

title = "The Susceptibility of Example-Based Explainability Methods to Class Outliers",

abstract = "This study explores the impact of class outliers on the effectiveness of example-based explainability methods for black-box machine learning models. We reformulate existing explainability evaluation metrics, such as correctness and relevance, specifically for example-based methods, and introduce a new metric, distinguishability. Using these metrics, we highlight the shortcomings of current example-based explainability methods, including those who attempt to suppress class outliers. We conduct experiments on two datasets, a text classification dataset and an image classification dataset, and evaluate the performance of four state-of-the-art explainability methods. Our findings underscore the need for robust techniques to tackle the challenges posed by class outliers.",

keywords = "explainability, explainability evaluation, interpretability",

author = "Ikhtiyor Nematov and Dimitris Sacharidis and Katja Hose and Tomer Sagi",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.; 24th Joint European Conference on Machine Learning and Knowledge Discovery in Databases, ECML PKDD 2024 ; Conference date: 09-09-2024 Through 13-09-2024",

year = "2026",

doi = "10.1007/978-3-032-25305-7\_2",

language = "English",

isbn = "9783032253040",

series = "Communications in Computer and Information Science",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "19--26",

editor = "Mattia Cerrato and Danguole Kalinauskaite and Mantas Luko{\v s}evicius and Kristina {\v S}utiene and Mykola Pechenizkiy",

booktitle = "Machine Learning and Principles and Practice of Knowledge Discovery in Databases - International Workshops of ECML PKDD 2024, Revised Selected Papers",

address = "Germany",

}

Nematov, I, Sacharidis, D, Hose, K & Sagi, T 2026, The Susceptibility of Example-Based Explainability Methods to Class Outliers. in M Cerrato, D Kalinauskaite, M Lukoševicius, K Šutiene & M Pechenizkiy (eds), Machine Learning and Principles and Practice of Knowledge Discovery in Databases - International Workshops of ECML PKDD 2024, Revised Selected Papers. Communications in Computer and Information Science, vol. 2559 CCIS, Springer Science and Business Media Deutschland GmbH, pp. 19-26, 24th Joint European Conference on Machine Learning and Knowledge Discovery in Databases, ECML PKDD 2024, Vilnius, Lithuania, 9/09/24. https://doi.org/10.1007/978-3-032-25305-7_2

The Susceptibility of Example-Based Explainability Methods to Class Outliers. / Nematov, Ikhtiyor; Sacharidis, Dimitris; Hose, Katja et al.
Machine Learning and Principles and Practice of Knowledge Discovery in Databases - International Workshops of ECML PKDD 2024, Revised Selected Papers. ed. / Mattia Cerrato; Danguole Kalinauskaite; Mantas Lukoševicius; Kristina Šutiene; Mykola Pechenizkiy. Springer Science and Business Media Deutschland GmbH, 2026. p. 19-26 (Communications in Computer and Information Science; Vol. 2559 CCIS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - The Susceptibility of Example-Based Explainability Methods to Class Outliers

AU - Nematov, Ikhtiyor

AU - Sacharidis, Dimitris

AU - Hose, Katja

AU - Sagi, Tomer

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

PY - 2026

Y1 - 2026

N2 - This study explores the impact of class outliers on the effectiveness of example-based explainability methods for black-box machine learning models. We reformulate existing explainability evaluation metrics, such as correctness and relevance, specifically for example-based methods, and introduce a new metric, distinguishability. Using these metrics, we highlight the shortcomings of current example-based explainability methods, including those who attempt to suppress class outliers. We conduct experiments on two datasets, a text classification dataset and an image classification dataset, and evaluate the performance of four state-of-the-art explainability methods. Our findings underscore the need for robust techniques to tackle the challenges posed by class outliers.

AB - This study explores the impact of class outliers on the effectiveness of example-based explainability methods for black-box machine learning models. We reformulate existing explainability evaluation metrics, such as correctness and relevance, specifically for example-based methods, and introduce a new metric, distinguishability. Using these metrics, we highlight the shortcomings of current example-based explainability methods, including those who attempt to suppress class outliers. We conduct experiments on two datasets, a text classification dataset and an image classification dataset, and evaluate the performance of four state-of-the-art explainability methods. Our findings underscore the need for robust techniques to tackle the challenges posed by class outliers.

KW - explainability

KW - explainability evaluation

KW - interpretability

UR - https://www.scopus.com/pages/publications/105040213684

U2 - 10.1007/978-3-032-25305-7_2

DO - 10.1007/978-3-032-25305-7_2

M3 - Conference contribution

AN - SCOPUS:105040213684

SN - 9783032253040

T3 - Communications in Computer and Information Science

SP - 19

EP - 26

BT - Machine Learning and Principles and Practice of Knowledge Discovery in Databases - International Workshops of ECML PKDD 2024, Revised Selected Papers

A2 - Cerrato, Mattia

A2 - Kalinauskaite, Danguole

A2 - Lukoševicius, Mantas

A2 - Šutiene, Kristina

A2 - Pechenizkiy, Mykola

PB - Springer Science and Business Media Deutschland GmbH

T2 - 24th Joint European Conference on Machine Learning and Knowledge Discovery in Databases, ECML PKDD 2024

Y2 - 9 September 2024 through 13 September 2024

ER -

Nematov I, Sacharidis D, Hose K, Sagi T. The Susceptibility of Example-Based Explainability Methods to Class Outliers. In Cerrato M, Kalinauskaite D, Lukoševicius M, Šutiene K, Pechenizkiy M, editors, Machine Learning and Principles and Practice of Knowledge Discovery in Databases - International Workshops of ECML PKDD 2024, Revised Selected Papers. Springer Science and Business Media Deutschland GmbH. 2026. p. 19-26. (Communications in Computer and Information Science). doi: 10.1007/978-3-032-25305-7_2