Techniques for discrimination-free predictive models

F. Kamiran, T.G.K. Calders, M. Pechenizkiy

Research output: Chapter in Book/Report/Conference proceedingChapterAcademic

7 Citations (Scopus)
4 Downloads (Pure)

Abstract

In this chapter, we give an overview of the techniques developed ourselves for constructing discrimination-free classifiers. In discrimination-free classification the goal is to learn a predictive model that classifies future data objects as accurately as possible, yet the predicted labels should be uncorrelated to a given sensitive attribute. For example, the task could be to learn a gender-neutral model that predicts whether a potential client of a bank has a high income or not. The techniques we developed for discrimination-aware classification can be divided into three categories: (1) removing the discrimination directly from the historical dataset before an off-the-shelf classification technique is applied; (2) changing the learning procedures themselves by restricting the search space to non-discriminatory models; and (3) adjusting the discriminatory models, learnt by off-the-shelf classifiers on discriminatory historical data, in a post-processing phase. Experiments show that even with such a strong constraint as discrimination-freeness, still very accurate models can be learnt. In particular,we study a case of income prediction,where the available historical data exhibits a wage gap between the genders. Due to legal restrictions, however, our predictions should be gender-neutral. The discrimination-aware techniques succeed in significantly reducing gender discrimination without impairing too much the accuracy.
Original languageEnglish
Title of host publicationDiscrimination and Privacy in the Information Society: Effects of Data Mining and Profiling Large Databases
EditorsB.H.M. Custers, T.G.K. Calders, B.W. Schermer, T.Z. Zarsky
Place of PublicationBerlin
PublisherSpringer
Chapter12
Pages223-239
ISBN (Print)978-3-642-30486-6
DOIs
Publication statusPublished - 2013

Publication series

NameStudies in Applied Philosophy, Epistemology and Rational Ethics
Volume3
ISSN (Print)2192-6255

Fingerprint

Classifiers
Wages
Labels
Processing
Experiments

Cite this

Kamiran, F., Calders, T. G. K., & Pechenizkiy, M. (2013). Techniques for discrimination-free predictive models. In B. H. M. Custers, T. G. K. Calders, B. W. Schermer, & T. Z. Zarsky (Eds.), Discrimination and Privacy in the Information Society: Effects of Data Mining and Profiling Large Databases (pp. 223-239). (Studies in Applied Philosophy, Epistemology and Rational Ethics; Vol. 3). Berlin: Springer. https://doi.org/10.1007/978-3-642-30487-3_12
Kamiran, F. ; Calders, T.G.K. ; Pechenizkiy, M. / Techniques for discrimination-free predictive models. Discrimination and Privacy in the Information Society: Effects of Data Mining and Profiling Large Databases. editor / B.H.M. Custers ; T.G.K. Calders ; B.W. Schermer ; T.Z. Zarsky. Berlin : Springer, 2013. pp. 223-239 (Studies in Applied Philosophy, Epistemology and Rational Ethics).
@inbook{4563159fd07740148e23b765bfba0ed4,
title = "Techniques for discrimination-free predictive models",
abstract = "In this chapter, we give an overview of the techniques developed ourselves for constructing discrimination-free classifiers. In discrimination-free classification the goal is to learn a predictive model that classifies future data objects as accurately as possible, yet the predicted labels should be uncorrelated to a given sensitive attribute. For example, the task could be to learn a gender-neutral model that predicts whether a potential client of a bank has a high income or not. The techniques we developed for discrimination-aware classification can be divided into three categories: (1) removing the discrimination directly from the historical dataset before an off-the-shelf classification technique is applied; (2) changing the learning procedures themselves by restricting the search space to non-discriminatory models; and (3) adjusting the discriminatory models, learnt by off-the-shelf classifiers on discriminatory historical data, in a post-processing phase. Experiments show that even with such a strong constraint as discrimination-freeness, still very accurate models can be learnt. In particular,we study a case of income prediction,where the available historical data exhibits a wage gap between the genders. Due to legal restrictions, however, our predictions should be gender-neutral. The discrimination-aware techniques succeed in significantly reducing gender discrimination without impairing too much the accuracy.",
author = "F. Kamiran and T.G.K. Calders and M. Pechenizkiy",
year = "2013",
doi = "10.1007/978-3-642-30487-3_12",
language = "English",
isbn = "978-3-642-30486-6",
series = "Studies in Applied Philosophy, Epistemology and Rational Ethics",
publisher = "Springer",
pages = "223--239",
editor = "B.H.M. Custers and T.G.K. Calders and B.W. Schermer and T.Z. Zarsky",
booktitle = "Discrimination and Privacy in the Information Society: Effects of Data Mining and Profiling Large Databases",
address = "Germany",

}

Kamiran, F, Calders, TGK & Pechenizkiy, M 2013, Techniques for discrimination-free predictive models. in BHM Custers, TGK Calders, BW Schermer & TZ Zarsky (eds), Discrimination and Privacy in the Information Society: Effects of Data Mining and Profiling Large Databases. Studies in Applied Philosophy, Epistemology and Rational Ethics, vol. 3, Springer, Berlin, pp. 223-239. https://doi.org/10.1007/978-3-642-30487-3_12

Techniques for discrimination-free predictive models. / Kamiran, F.; Calders, T.G.K.; Pechenizkiy, M.

Discrimination and Privacy in the Information Society: Effects of Data Mining and Profiling Large Databases. ed. / B.H.M. Custers; T.G.K. Calders; B.W. Schermer; T.Z. Zarsky. Berlin : Springer, 2013. p. 223-239 (Studies in Applied Philosophy, Epistemology and Rational Ethics; Vol. 3).

Research output: Chapter in Book/Report/Conference proceedingChapterAcademic

TY - CHAP

T1 - Techniques for discrimination-free predictive models

AU - Kamiran, F.

AU - Calders, T.G.K.

AU - Pechenizkiy, M.

PY - 2013

Y1 - 2013

N2 - In this chapter, we give an overview of the techniques developed ourselves for constructing discrimination-free classifiers. In discrimination-free classification the goal is to learn a predictive model that classifies future data objects as accurately as possible, yet the predicted labels should be uncorrelated to a given sensitive attribute. For example, the task could be to learn a gender-neutral model that predicts whether a potential client of a bank has a high income or not. The techniques we developed for discrimination-aware classification can be divided into three categories: (1) removing the discrimination directly from the historical dataset before an off-the-shelf classification technique is applied; (2) changing the learning procedures themselves by restricting the search space to non-discriminatory models; and (3) adjusting the discriminatory models, learnt by off-the-shelf classifiers on discriminatory historical data, in a post-processing phase. Experiments show that even with such a strong constraint as discrimination-freeness, still very accurate models can be learnt. In particular,we study a case of income prediction,where the available historical data exhibits a wage gap between the genders. Due to legal restrictions, however, our predictions should be gender-neutral. The discrimination-aware techniques succeed in significantly reducing gender discrimination without impairing too much the accuracy.

AB - In this chapter, we give an overview of the techniques developed ourselves for constructing discrimination-free classifiers. In discrimination-free classification the goal is to learn a predictive model that classifies future data objects as accurately as possible, yet the predicted labels should be uncorrelated to a given sensitive attribute. For example, the task could be to learn a gender-neutral model that predicts whether a potential client of a bank has a high income or not. The techniques we developed for discrimination-aware classification can be divided into three categories: (1) removing the discrimination directly from the historical dataset before an off-the-shelf classification technique is applied; (2) changing the learning procedures themselves by restricting the search space to non-discriminatory models; and (3) adjusting the discriminatory models, learnt by off-the-shelf classifiers on discriminatory historical data, in a post-processing phase. Experiments show that even with such a strong constraint as discrimination-freeness, still very accurate models can be learnt. In particular,we study a case of income prediction,where the available historical data exhibits a wage gap between the genders. Due to legal restrictions, however, our predictions should be gender-neutral. The discrimination-aware techniques succeed in significantly reducing gender discrimination without impairing too much the accuracy.

U2 - 10.1007/978-3-642-30487-3_12

DO - 10.1007/978-3-642-30487-3_12

M3 - Chapter

SN - 978-3-642-30486-6

T3 - Studies in Applied Philosophy, Epistemology and Rational Ethics

SP - 223

EP - 239

BT - Discrimination and Privacy in the Information Society: Effects of Data Mining and Profiling Large Databases

A2 - Custers, B.H.M.

A2 - Calders, T.G.K.

A2 - Schermer, B.W.

A2 - Zarsky, T.Z.

PB - Springer

CY - Berlin

ER -

Kamiran F, Calders TGK, Pechenizkiy M. Techniques for discrimination-free predictive models. In Custers BHM, Calders TGK, Schermer BW, Zarsky TZ, editors, Discrimination and Privacy in the Information Society: Effects of Data Mining and Profiling Large Databases. Berlin: Springer. 2013. p. 223-239. (Studies in Applied Philosophy, Epistemology and Rational Ethics). https://doi.org/10.1007/978-3-642-30487-3_12