Abstract
iIn several previous studies, accuracy of K-Nearest Neighbor (KNN) iobtained a less ithan maximum accuracy value when compared to other methods. iAs for the causes of ithis, one of which is ibecause ieach iattribute ihas ithe isame iimpact ion ithe iclassification iprocess, iwhile isome characteristics that are iless irelevant cause misclassification in idetermining ithe iclass ifor inew idata. Then iin ithis istudy, the author proposes to K-iNearest Neighbor ithat iattribute iselection ibe icarried iout ito iremove iless irelevant iattributes ibefore icarrying iout ithe iclassification iprocess. iThe iattribute iselection imethod iused iin ithis istudy iis ithe iRelief-F iAlgorithm ias ia imethod ifor attribute iselection whose correlation is not igood ifrom ithe idataset ibeing itested. iThe accuracy iresults iobtained iwill ibe icompared iwith ithe iaccuracy iobtained ifrom ithe iconventional iKNN method using iConfusion Matrix. iThe itest iresults iobtained iare ithat ithe iproposed imethod iis iable ito iincrease ithe iclassification iaccuracy iof iKNN, iwhere ithe iincrease iin iaccuracy iobtained iafter ithe iattribute iselection iis i2.02%.
