Techniques for Discrimination-Free Predictive Models - Discrimination and Privacy in the Information Society - page 229

we assume there is just one sensitive attribute, dividing the objects into one disad-
vantaged and one advantaged group. Often, however, there may be more than two
groups, each of which are advantaged/disadvantaged to a different level. Consider,
e.g., different ethnic minorities being treated in different ways. Furthermore, there
may be multiple of such sensitive attributes; e.g., gender, age, and ethnicity. Remov-
ing gender-discrimination by the preprocessing techniques may introduce an age-
discrimination. Furthermore, it could be the case that even if discrimination does
not manifest itself at the general level, in some specialized niches or contexts, there
might be discrimination present. Chapter 5 of this topic deals with the detection of
such subtle contexts for discrimination. Also, as discussed in Chapter 8 of this topic,
not all difference in acceptance rates between an advantaged and a disadvantaged
group is due to discrimination. If people in the disadvantaged group are more likely
to be lowly educated, as a result their salaries will be lower on average, without
this difference necessarily indicating a discrimination. As a conclusion, the area of
discrimination-aware classification remains a rich source of inspiration and applica-
tion area for novel techniques in the data mining area, and we hope to see significant
contributions in future to this ethically and societally important research area, lead-
ing towards providing companies and practitioners with the necessary toolkit for
data-driven discrimination-free decision making.
References
Australian Law. Australian Sex Discimination Act 1984. Australian sex discimination act
1984. via: (1984) http://www.comlaw.gov.au/Details/C2010C00056
Calders, T., Kamiran, F., Pechenizkiy, M.: Building Classifiers with Independency Con-
straintsBuilding classifiers with independency constraints. In: Saygin, Y., et al. (eds.)
ICDM Workshops 2009, IEEE International Conference on Data Mining Workshops,
Miami, Florida, USA, December 6, pp. 13-18 (2009)
Calders, T., Verwer, S.: Three naive Bayes approaches for discrimination-free classifica-
tionThree naive bayes approaches for discrimination-free classification. Data Mining and
Knowledge Discovery 21, 277-292 (2010)
Frank, A., Asuncion, A.: UCI Machine Learning Repository. UCI machine learning reposi-
tory (2010), http://archive.ics.uci.edu/ml
Hajian, S., Domingo-Ferrer, J., Martinez-Balleste, A.: Discrimination prevention in data min-
ing for intrusion and crime detectionDiscrimination prevention in data mining for intrusion
and crime detection. In: IEEE Symposium on Computational Intelligence in Cyber Secu-
rity (CICS)IEEE Symposium on Computational Intelligence in Cyber Security (CICS),
pp. 47-54 (2011)
Hajian, S., Domingo-Ferrer, J., Martınez-Balleste, A.: Rule protection for indirect discrimi-
nation prevention in data miningRule protection for indirect discrimination prevention in
data mining. In: Modeling Decision for Artificial Intelligence, pp. 211-222 (2011)
Hall, M., Frank, E., Holmes, G., Pfahringer, B., Reutemann, P., Witten, I.: The WEKA data
mining software: An updateThe WEKA data mining software: An update. ACM SIGKDD
Explorations Newsletter 11, 110-118 (2009)