AutoXplain: Towards Automated Interpretable Model Selection

Tessel Haagen; Heysem Kaya; Joop Snijder; Melchior Nierman

AutoXplain: Towards Automated Interpretable Model Selection

Tessel Haagen^*
, Heysem Kaya
, Joop Snijder
, Melchior Nierman

^*Corresponding author for this work

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

Abstract

Machine learning (ML) algorithms are increasingly used in high-stake domains like healthcare. While ML systems frequently outperform humans in specific tasks, ensuring safety and transparency is critical in these domains. Interpretability, therefore, plays a crucial role in understanding the decision-making process, auditing and correction of ML models and establishing trust. Furthermore, there is a growing demand for automated machine learning (AutoML) to facilitate model development without expert intervention. However, the combination of interpretability and AutoML has received limited attention
thus far. In this study, we propose two objective model-agnostic measures of interpretability to quantify model compactness and explanation stability, embedded within an automated interpretable ML pipeline.
We experiment with a set of interpretable models on medical classification tasks reporting the proposed measures along with the predictive performances. We further conduct a user study with domain experts to evaluate the correlation between these measures and the subjective concept of interpretability. Our
findings demonstrate the effectiveness of the proposed measures, affirming their success and validating their utility in creating an interpretable automated pipeline.

Original language	English
Title of host publication	xAI-2023 Late-breaking Work, Demos and Doctoral Consortium Joint Proceedings
Subtitle of host publication	Joint Proceedings of the xAI-2023 Late-breaking Work, Demos and Doctoral Consortium co-located with the 1st World Conference on eXplainable Artificial Intelligence (xAI-2023)
Editors	Luca Longo
Publisher	CEUR WS
Pages	18-23
Number of pages	6
Volume	3554
Publication status	Published - 20 Nov 2023
Event	The 1st World Conference on eXplainable Artificial Intelligence - Lisbon, Lisboa, Portugal Duration: 26 Jul 2022 → 28 Jul 2022 Conference number: 1 https://xaiworldconference.com/2023/

Publication series

Name	CEUR Workshop Proceedings
ISSN (Print)	1613-0073

Conference

Conference	The 1st World Conference on eXplainable Artificial Intelligence
Abbreviated title	xAI 2023
Country/Territory	Portugal
City	Lisboa
Period	26/07/22 → 28/07/22
Internet address	https://xaiworldconference.com/2023/

Bibliographical note

Publisher Copyright:
© 2023 CEUR-WS. All rights reserved.

Funding

The research leading to this publication is conducted by T. Haagen while she was pursuing MSc thesis internship at InfoSupport company on the Atalmedial dataset.

Keywords

Automated Machine Learning (AutoML)
Interpretability measures
Interpretable automated pipeline
Machine Learning for health-care
Model-agnostic measures

Access to Document

paper4Final published version, 1.01 MBLicence: CC BY

https://ceur-ws.org/Vol-3554/paper4.pdfLicence: CC BY

Cite this

Haagen, T., Kaya, H., Snijder, J., & Nierman, M. (2023). AutoXplain: Towards Automated Interpretable Model Selection. In L. Longo (Ed.), xAI-2023 Late-breaking Work, Demos and Doctoral Consortium Joint Proceedings: Joint Proceedings of the xAI-2023 Late-breaking Work, Demos and Doctoral Consortium co-located with the 1st World Conference on eXplainable Artificial Intelligence (xAI-2023) (Vol. 3554, pp. 18-23). (CEUR Workshop Proceedings). CEUR WS. https://ceur-ws.org/Vol-3554/paper4.pdf

Haagen, Tessel ; Kaya, Heysem ; Snijder, Joop et al. / AutoXplain: Towards Automated Interpretable Model Selection. xAI-2023 Late-breaking Work, Demos and Doctoral Consortium Joint Proceedings: Joint Proceedings of the xAI-2023 Late-breaking Work, Demos and Doctoral Consortium co-located with the 1st World Conference on eXplainable Artificial Intelligence (xAI-2023). editor / Luca Longo. Vol. 3554 CEUR WS, 2023. pp. 18-23 (CEUR Workshop Proceedings).

@inproceedings{bab3ea92d4ab48ce814a0e197b00d442,

title = "AutoXplain: Towards Automated Interpretable Model Selection",

abstract = "Machine learning (ML) algorithms are increasingly used in high-stake domains like healthcare. While ML systems frequently outperform humans in specific tasks, ensuring safety and transparency is critical in these domains. Interpretability, therefore, plays a crucial role in understanding the decision-making process, auditing and correction of ML models and establishing trust. Furthermore, there is a growing demand for automated machine learning (AutoML) to facilitate model development without expert intervention. However, the combination of interpretability and AutoML has received limited attention thus far. In this study, we propose two objective model-agnostic measures of interpretability to quantify model compactness and explanation stability, embedded within an automated interpretable ML pipeline. We experiment with a set of interpretable models on medical classification tasks reporting the proposed measures along with the predictive performances. We further conduct a user study with domain experts to evaluate the correlation between these measures and the subjective concept of interpretability. Our findings demonstrate the effectiveness of the proposed measures, affirming their success and validating their utility in creating an interpretable automated pipeline.",

keywords = "Automated Machine Learning (AutoML), Interpretability measures, Interpretable automated pipeline, Machine Learning for health-care, Model-agnostic measures",

author = "Tessel Haagen and Heysem Kaya and Joop Snijder and Melchior Nierman",

note = "Publisher Copyright: {\textcopyright} 2023 CEUR-WS. All rights reserved.; The 1st World Conference on eXplainable Artificial Intelligence, xAI 2023 ; Conference date: 26-07-2022 Through 28-07-2022",

year = "2023",

month = nov,

day = "20",

language = "English",

volume = "3554",

series = "CEUR Workshop Proceedings",

publisher = "CEUR WS",

pages = "18--23",

editor = "Longo, \{Luca \}",

booktitle = "xAI-2023 Late-breaking Work, Demos and Doctoral Consortium Joint Proceedings",

url = "https://xaiworldconference.com/2023/",

}

Haagen, T, Kaya, H, Snijder, J & Nierman, M 2023, AutoXplain: Towards Automated Interpretable Model Selection. in L Longo (ed.), xAI-2023 Late-breaking Work, Demos and Doctoral Consortium Joint Proceedings: Joint Proceedings of the xAI-2023 Late-breaking Work, Demos and Doctoral Consortium co-located with the 1st World Conference on eXplainable Artificial Intelligence (xAI-2023). vol. 3554, CEUR Workshop Proceedings, CEUR WS, pp. 18-23, The 1st World Conference on eXplainable Artificial Intelligence, Lisboa, Portugal, 26/07/22. <https://ceur-ws.org/Vol-3554/paper4.pdf>

AutoXplain: Towards Automated Interpretable Model Selection. / Haagen, Tessel; Kaya, Heysem; Snijder, Joop et al.
xAI-2023 Late-breaking Work, Demos and Doctoral Consortium Joint Proceedings: Joint Proceedings of the xAI-2023 Late-breaking Work, Demos and Doctoral Consortium co-located with the 1st World Conference on eXplainable Artificial Intelligence (xAI-2023). ed. / Luca Longo. Vol. 3554 CEUR WS, 2023. p. 18-23 (CEUR Workshop Proceedings).

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

TY - GEN

T1 - AutoXplain: Towards Automated Interpretable Model Selection

AU - Haagen, Tessel

AU - Kaya, Heysem

AU - Snijder, Joop

AU - Nierman, Melchior

N1 - Conference code: 1

PY - 2023/11/20

Y1 - 2023/11/20

N2 - Machine learning (ML) algorithms are increasingly used in high-stake domains like healthcare. While ML systems frequently outperform humans in specific tasks, ensuring safety and transparency is critical in these domains. Interpretability, therefore, plays a crucial role in understanding the decision-making process, auditing and correction of ML models and establishing trust. Furthermore, there is a growing demand for automated machine learning (AutoML) to facilitate model development without expert intervention. However, the combination of interpretability and AutoML has received limited attention thus far. In this study, we propose two objective model-agnostic measures of interpretability to quantify model compactness and explanation stability, embedded within an automated interpretable ML pipeline. We experiment with a set of interpretable models on medical classification tasks reporting the proposed measures along with the predictive performances. We further conduct a user study with domain experts to evaluate the correlation between these measures and the subjective concept of interpretability. Our findings demonstrate the effectiveness of the proposed measures, affirming their success and validating their utility in creating an interpretable automated pipeline.

AB - Machine learning (ML) algorithms are increasingly used in high-stake domains like healthcare. While ML systems frequently outperform humans in specific tasks, ensuring safety and transparency is critical in these domains. Interpretability, therefore, plays a crucial role in understanding the decision-making process, auditing and correction of ML models and establishing trust. Furthermore, there is a growing demand for automated machine learning (AutoML) to facilitate model development without expert intervention. However, the combination of interpretability and AutoML has received limited attention thus far. In this study, we propose two objective model-agnostic measures of interpretability to quantify model compactness and explanation stability, embedded within an automated interpretable ML pipeline. We experiment with a set of interpretable models on medical classification tasks reporting the proposed measures along with the predictive performances. We further conduct a user study with domain experts to evaluate the correlation between these measures and the subjective concept of interpretability. Our findings demonstrate the effectiveness of the proposed measures, affirming their success and validating their utility in creating an interpretable automated pipeline.

KW - Automated Machine Learning (AutoML)

KW - Interpretability measures

KW - Interpretable automated pipeline

KW - Machine Learning for health-care

KW - Model-agnostic measures

UR - https://www.scopus.com/pages/publications/85178603203

M3 - Conference contribution

VL - 3554

T3 - CEUR Workshop Proceedings

SP - 18

EP - 23

BT - xAI-2023 Late-breaking Work, Demos and Doctoral Consortium Joint Proceedings

A2 - Longo, Luca

PB - CEUR WS

T2 - The 1st World Conference on eXplainable Artificial Intelligence

Y2 - 26 July 2022 through 28 July 2022

ER -

Haagen T, Kaya H, Snijder J, Nierman M. AutoXplain: Towards Automated Interpretable Model Selection. In Longo L, editor, xAI-2023 Late-breaking Work, Demos and Doctoral Consortium Joint Proceedings: Joint Proceedings of the xAI-2023 Late-breaking Work, Demos and Doctoral Consortium co-located with the 1st World Conference on eXplainable Artificial Intelligence (xAI-2023). Vol. 3554. CEUR WS. 2023. p. 18-23. (CEUR Workshop Proceedings).

AutoXplain: Towards Automated Interpretable Model Selection

Abstract

Publication series

Conference

Bibliographical note

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this