The Discovery of Discrimination - Discrimination and Privacy in the Information Society - page 102

Indirect discrimination discovery , such as the following redlining question “I
don't have the race attribute in my data, but have the ZIP of residence. By adding
background knowledge on the distribution of race over ZIP codes, infer cases
where ZIP actually disguises race discrimination.”

Affirmative actions and favoritism: “List cases where our university admission
policies actually favored blacks” ,and “Under which conditions white males are
given the best mortgage rate in comparison to the average?”

On-line documentation, demo, and download of the DCUBE system can be accessed
from http://kdd.di.unipi.it/dcube .

5.8 Conclusions

We presented a data mining approach for the analysis and discovery of discrimi-
nation in a dataset of socially-sensitive decisions. The approach consists first of ex-
tracting frequent classification rules, and then screening/ranking them on the basis of
quantitative measures of discrimination. The key legal concepts of protected-by-law
groups, direct discrimination, indirect discrimination, genuine occupational require-
ment, and affirmative actions are formalized as reasonings over the set of extracted
rules and, possibly, additional background knowledge. The approach has been im-
plemented in the DCUBE tool and made publicly available. Chapter 13 builds on
our approach for the purpose of designing data mining classifiers that do not learn
to discriminate, an issue known as discrimination prevention.
As future work, we aim to achieve two goals: on one hand, to improve the meth-
ods and the technologies for discovering discrimination, especially looking at data
mining methods such at classification and clustering, driven by constraints over spe-
cific application contexts (racial profiling, labor market, credit scoring, etc.); on the
other hand, to further interact with legal experts both to find out new measures and
rules that we may support with our tools and to influence their design and inter-
pretation of legislation. Finally, we are looking at other fields of application, other
than credit scoring. An interesting one is discovering possible discrimination (with
respect to sex, nationality, etc.) in funding research projects.

References

Agrawal, R., Srikant, R.: Fast algorithms for mining association rules in large databases.
In: Proc. of Int. Conf. on Very Large Data Bases (VLDB 1994), pp. 487-499. Morgan
Kaufmann (1994)
Agrawal, R., Srikant, R.: Privacy-preserving data mining. In: Proc. of the ACM SIGMOD
Int. Conf. on Management of Data (SIGMOD 2000), pp. 439-450. ACM (2000)
Australian Legislation: (a) Age Discrimination Act, 2004; (b) Australian Human Rights Com-
mission Act, 1986; (c) Disability Discrimination Act, 1992; (d) Racial Discrimination Act,
1975; (e) Sex Discrimination Act, 1984; (f) Victoria Equal Opportunity Act, 1995; (g)
Queensland Anti Discrimination Act, 1991 (2011), http://www.hreoc.gov.au