Anti-discrimination, Data Mining,Direct and indirect discrimination prevention, Privacy, Rule generalization,Rule protection.
Abstract
Data mining is a progressively more important technology for extracting useful data from large amount of data. There are unfavorable social viewpoint about data mining, among which potential privacy intrusion and possible discrimination. The latter consists of unjustly dealing with people on the basis of their belonging to a peculiar group. Automatic data set and data mining approaches such as classification rule mining have covered the way to making computerized decisions, like loan accepting/denial. If the training data sets are limited in what regards discriminatory attributes like religion, gender, race, etc., discriminatory decisions may ensue. For this cause, antidiscrimination approaches including discrimination detection and prevention have been introduced in data mining. Discrimination can be either direct or indirect. Direct discrimination appears when judgments are made based on responsive attributes. Indirect discrimination appears when decisions are made based on no sensitive attributes which are strongly associated with prejudiced sensitive ones. In this paper, we tackle discrimination avoidance in data mining and recommend new techniques useful for direct or indirect discrimination prevention individually or both at the equivalent time. We confer how to clean teaching data sets and outsourced data sets in such a way that direct and/or indirect discriminatory decision rules are converted to legitimate (nondiscriminatory) classification rules. The experimental calculations reveal that the suggested techniques are impressive at removing direct and/or indirect discrimination biases in the initial data set while protecting data quality.
