Discriminatory Association Analysis on Semi-structured Data

Data mining has been applied to the discovery of illegally discriminatory treatments caused by protected-by-law attributes such as race, gender, age, etc. In this paper, we propose an improvement for the previous work of exploring discrimination in semi-structured business data. The main idea is that discrimination represented in the form of association rules is judged by opposite patterns whose components are almost the same except a single sensitive attribute and the decision (admission to school, acceptance to a position, etc.) However, the previous work requires numerous efforts and it has not been modeled in a systematic way. In order to solve this limitation, semantic analysis is integrated in the discrimination mining process showing better results in comparison with the previous work in experimental outcome.