Differentially private feature selection under MapReduce framework

(整期优先)网络出版时间:2013-05-15
/ 1
Privacypreservingdataminingalgorithmsarecrucialforthepersonaldataanalysis,suchasmedicalandfinancialrecords.Thispaperfocusesonfeatureselectionandproposesanewprivacypreservingdistributedalgorithm,whichcaneffectivelyselectfeaturesbasedondifferentialprivacyandGiniindexundertheMapReduceframework.Atthesametime,thetheoreticanalysisforprivacyguaranteeisalsopresented.Someexperimentsareconductedonbench-markdatasets,thesimulationresultsindicatethatduringtheselectionofimportantfeatures,theproposedalgorithmcanpreserveprivacyinformationtoacertainextentwithlesstimecostthanoncentralizedcounterpart.