Method for Preventing Direct and Indirect Discrimination in Data Mining

Abstract

For extorting the helpful comprehension concealed in the biggest compilation of a database the data mining technology is used. There are some negative approaches occurred about the data mining technology, among which the potential privacy incursion and potential discrimination. The latter consists of irrationally considering individuals on the source of their fitting to an exact group. Data mining and automated data collection methods like the classification covered the way forsaking the automated judgment like granting or denying the loan on the basis of race, creed, etc. If the training data sets are unfair in what respects discriminatory attributes like masculine category, race, creed, etc., discriminator decisions may ensue. Because of this reason the data mining technology introduced ant discrimination methods with including the discrimination discovery and avoidance. The discrimination can direct or indirect. When any decisions were made to the sensitive attributes at that time direct discrimination are occurring. While the indirect discrimination are occurring when the decision remade on the basis of non-sensitive attributes which are strongly associated with the sensitive. Here in this paper, we deal with discrimination avoidance in data mining and proposed novel method for discrimination prevention with the post processing approach. We projected Classification based on predictive association rules (CPAR) algorithm, which is a kind of association classification methods. The algorithm combines the advantages of both association classification methods and traditional rule based classification. The algorithm used to thwart discrimination deterrence in post processing. We calculate the utility of the proposed approach and compare with the existing approaches. The experimental assessment proved that the proposed method is effectively removing the direct or unintended discrimination prejudices in the original data set for maintaining the quality of data.