DSpace Collection:http://localhost:8080/xmlui/handle/123456789/29652026-06-23T06:21:35Z2026-06-23T06:21:35ZIM_LR: An approach for Direct and Indirect Discrimination PreventionWakchaure, M.A.Sane, S. S.http://localhost:8080/xmlui/handle/123456789/29662021-07-17T09:20:19Z2019-05-15T00:00:00ZTitle: IM_LR: An approach for Direct and Indirect Discrimination Prevention
Authors: Wakchaure, M.A.; Sane, S. S.
Abstract: Discrimination and privacy preservation are major
challenges of data mining. Technique based on impact
minimization to prevent discrimination has been reported in the
literature. The technique computes fitness of generated frequent
rules based on their antecedent, a pre-defined threshold and
discrimination measure ‘elift’ to modify discriminating rules. This
paper deals with a method called ‘IMLR’. IMLR computes fitness
of generated frequent rules based on their antecedent (attributes
on left hand side of the rule) as well as consequences (class label
on right hand side of the rule), a pre-defined threshold and offers
selection of desired discrimination measures such as ‘elift’, ‘slift’,
‘olift’ etc. to modify discriminating rules. Experimentation results
carried out using two well-known datasets ‘Adult’ and ‘German’
show that IMLR when used with certain discrimination measure
provides better results in terms of various performance parameters
such as DDPD, DDPP, IDPD, IDPP, Missed cost and Ghost cost
when compared with reported technique2019-05-15T00:00:00Z