Preprocessing Matters: Automated Pipeline Selection for Fair Classification

Vladimiro González-Zelaya; Julián Salas; Dennis Prangle; Paolo Missier

doi:10.1007/978-3-031-33498-6_14

Preprocessing Matters: Automated Pipeline Selection for Fair Classification

Vladimiro González-Zelaya, Julián Salas^*, Dennis Prangle, Paolo Missier

^*Corresponding author for this work

Computer Science

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

Abstract

Improving fairness by manipulating the preprocessing stages of classification pipelines is an active area of research, closely related to AutoML. We propose a genetic optimisation algorithm, FairPipes, which optimises for user-defined combinations of fairness and accuracy and for multiple definitions of fairness, providing flexibility in the fairness-accuracy trade-off. FairPipes heuristically searches through a large space of pipeline configurations, achieving near-optimality efficiently, presenting the user with an estimate of the solutions’ Pareto front. We also observe that the optimal pipelines differ for different datasets, suggesting that no “universal best” pipeline exists and confirming that FairPipes fills a niche in the fairness-aware AutoML space.

Original language	English
Title of host publication	Modeling Decisions for Artificial Intelligence - 20th International Conference, MDAI 2023, Proceedings
Editors	Vicenç Torra, Yasuo Narukawa
Publisher	Springer
Pages	202-213
Number of pages	12
ISBN (Print)	9783031334979
DOIs	https://doi.org/10.1007/978-3-031-33498-6_14
Publication status	Published - 19 May 2023
Event	20th International Conference on Modeling Decisions for Artificial Intelligence, MDAI 2023 - Umeå, Sweden Duration: 19 Jun 2023 → 22 Jun 2023

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	13890 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	20th International Conference on Modeling Decisions for Artificial Intelligence, MDAI 2023
Country/Territory	Sweden
City	Umeå
Period	19/06/23 → 22/06/23

Bibliographical note

Publisher Copyright:
© 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

Keywords

Algorithmic Fairness
AutoML
Data Preprocessing
Ethical AI
Genetic Algorithms
Preprocessing Pipelines

ASJC Scopus subject areas

Theoretical Computer Science
General Computer Science

Access to Document

10.1007/978-3-031-33498-6_14Licence: None: All rights reserved

Cite this

González-Zelaya, V., Salas, J., Prangle, D., & Missier, P. (2023). Preprocessing Matters: Automated Pipeline Selection for Fair Classification. In V. Torra, & Y. Narukawa (Eds.), Modeling Decisions for Artificial Intelligence - 20th International Conference, MDAI 2023, Proceedings (pp. 202-213). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13890 LNCS). Springer. https://doi.org/10.1007/978-3-031-33498-6_14

González-Zelaya, Vladimiro ; Salas, Julián ; Prangle, Dennis et al. / Preprocessing Matters : Automated Pipeline Selection for Fair Classification. Modeling Decisions for Artificial Intelligence - 20th International Conference, MDAI 2023, Proceedings. editor / Vicenç Torra ; Yasuo Narukawa. Springer, 2023. pp. 202-213 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{e2134c1ad6db40feb07061fb56aa8218,

title = "Preprocessing Matters: Automated Pipeline Selection for Fair Classification",

abstract = "Improving fairness by manipulating the preprocessing stages of classification pipelines is an active area of research, closely related to AutoML. We propose a genetic optimisation algorithm, FairPipes, which optimises for user-defined combinations of fairness and accuracy and for multiple definitions of fairness, providing flexibility in the fairness-accuracy trade-off. FairPipes heuristically searches through a large space of pipeline configurations, achieving near-optimality efficiently, presenting the user with an estimate of the solutions{\textquoteright} Pareto front. We also observe that the optimal pipelines differ for different datasets, suggesting that no “universal best” pipeline exists and confirming that FairPipes fills a niche in the fairness-aware AutoML space.",

keywords = "Algorithmic Fairness, AutoML, Data Preprocessing, Ethical AI, Genetic Algorithms, Preprocessing Pipelines",

author = "Vladimiro Gonz{\'a}lez-Zelaya and Juli{\'a}n Salas and Dennis Prangle and Paolo Missier",

note = "Publisher Copyright: {\textcopyright} 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 20th International Conference on Modeling Decisions for Artificial Intelligence, MDAI 2023 ; Conference date: 19-06-2023 Through 22-06-2023",

year = "2023",

month = may,

day = "19",

doi = "10.1007/978-3-031-33498-6_14",

language = "English",

isbn = "9783031334979",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer",

pages = "202--213",

editor = "Vicen{\c c} Torra and Yasuo Narukawa",

booktitle = "Modeling Decisions for Artificial Intelligence - 20th International Conference, MDAI 2023, Proceedings",

}

González-Zelaya, V, Salas, J, Prangle, D & Missier, P 2023, Preprocessing Matters: Automated Pipeline Selection for Fair Classification. in V Torra & Y Narukawa (eds), Modeling Decisions for Artificial Intelligence - 20th International Conference, MDAI 2023, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13890 LNCS, Springer, pp. 202-213, 20th International Conference on Modeling Decisions for Artificial Intelligence, MDAI 2023, Umeå, Sweden, 19/06/23. https://doi.org/10.1007/978-3-031-33498-6_14

Preprocessing Matters: Automated Pipeline Selection for Fair Classification. / González-Zelaya, Vladimiro; Salas, Julián; Prangle, Dennis et al.
Modeling Decisions for Artificial Intelligence - 20th International Conference, MDAI 2023, Proceedings. ed. / Vicenç Torra; Yasuo Narukawa. Springer, 2023. p. 202-213 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13890 LNCS).

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

TY - GEN

T1 - Preprocessing Matters

T2 - 20th International Conference on Modeling Decisions for Artificial Intelligence, MDAI 2023

AU - González-Zelaya, Vladimiro

AU - Salas, Julián

AU - Prangle, Dennis

AU - Missier, Paolo

PY - 2023/5/19

Y1 - 2023/5/19

N2 - Improving fairness by manipulating the preprocessing stages of classification pipelines is an active area of research, closely related to AutoML. We propose a genetic optimisation algorithm, FairPipes, which optimises for user-defined combinations of fairness and accuracy and for multiple definitions of fairness, providing flexibility in the fairness-accuracy trade-off. FairPipes heuristically searches through a large space of pipeline configurations, achieving near-optimality efficiently, presenting the user with an estimate of the solutions’ Pareto front. We also observe that the optimal pipelines differ for different datasets, suggesting that no “universal best” pipeline exists and confirming that FairPipes fills a niche in the fairness-aware AutoML space.

AB - Improving fairness by manipulating the preprocessing stages of classification pipelines is an active area of research, closely related to AutoML. We propose a genetic optimisation algorithm, FairPipes, which optimises for user-defined combinations of fairness and accuracy and for multiple definitions of fairness, providing flexibility in the fairness-accuracy trade-off. FairPipes heuristically searches through a large space of pipeline configurations, achieving near-optimality efficiently, presenting the user with an estimate of the solutions’ Pareto front. We also observe that the optimal pipelines differ for different datasets, suggesting that no “universal best” pipeline exists and confirming that FairPipes fills a niche in the fairness-aware AutoML space.

KW - Algorithmic Fairness

KW - AutoML

KW - Data Preprocessing

KW - Ethical AI

KW - Genetic Algorithms

KW - Preprocessing Pipelines

UR - http://www.scopus.com/inward/record.url?scp=85161217164&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-33498-6_14

DO - 10.1007/978-3-031-33498-6_14

M3 - Conference contribution

AN - SCOPUS:85161217164

SN - 9783031334979

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 202

EP - 213

BT - Modeling Decisions for Artificial Intelligence - 20th International Conference, MDAI 2023, Proceedings

A2 - Torra, Vicenç

A2 - Narukawa, Yasuo

PB - Springer

Y2 - 19 June 2023 through 22 June 2023

ER -

González-Zelaya V, Salas J, Prangle D, Missier P. Preprocessing Matters: Automated Pipeline Selection for Fair Classification. In Torra V, Narukawa Y, editors, Modeling Decisions for Artificial Intelligence - 20th International Conference, MDAI 2023, Proceedings. Springer. 2023. p. 202-213. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-33498-6_14