Optimasi Kinerja K-Means Clustering Menggunakan Pembobotan Symmetrical Uncertainty dalam Klasterisasi Data

Abstract

In some previous research, iK-Means iClustering ihas iseveral iweaknesses, ione of which ilies iin ithe idistance imodel iused in idetermining ithe isimilarity ibetween idata that igives ithe isame itreatment ito ieach idata iattribute, so that iattributes ithat iare iless relevant iand ihave ilittle icontribution ito idata ivariation ican iprovide isignificant iimpact on the results iof iclustering. This iof icourse ican ireduce ithe iperformance iof K-Means Clustering. Attribute iweighting is one iway ithat ican ibe iused ito iget ithe icorrelation of data iattributes ito idata ivariations. The ihigher the iweight ivalue of ian iattribute, the greater ithe icorrelation ito the ivariation iof the idata, so that ithe ilow iweight ivalue of an iattribute icertainly ihas ia iismall icontribution to the variation of the data and can have ia isignificant iimpact ion ithe iperformance iand iresults of iclustering. In ithis istudy, the imethod iused in icalculating ithe iweight of data attributes is Symmetrical Uncertainty. To test the proposed method, this research uses a dataset from UCI Machine Learning which consists of Iris with 150 data and Wine Quality with 178 data. The ievaluation iof ithe iproposed iclustering iperformance iis ibased ion ithe iDavies-Bouldin iIndex (DBI) ivalue. The itest iresults iin ithis istudy ishow ithat the iproposed method ican iproduce ia isignificantly ilarger Davies-Bouldin Index (DBI) value

ikelemahan, salah isatunya iterletak ipada imodel ijarak iyang idigunakan idalam ipenentuan ikemiripan antar idata iyang imemberikan iperlakuan iyang isama iterhadap isetiap iatribut idata, sehingga iatribut iyang ikurang irelevan idan imemiliki isedikit ikontribusi iterhadap variasi idata idapat imemberikan idampak iyang icukup iberpengaruh iterhadap ihasil clustering. iHal iini itentu isaja idapat imenurunkan ikinerja iK-Means iClustering. Pembobotan iatribut imerupakan isalah isatu icara iyang idapat idigunakan iuntuk mendapatkan ikorelasi iatribut idata iterhadap ivariasi idata. iSemakin itinggi inilai ibobot dari isuatu iatribut imaka isemakin ibesar ikorelasinya iterhadap ivariasi idata, isehingga nilai ibobot iyang irendah idari isuatu iatribut itentunya imemiliki isedikit ikontribusi terhadap ivariasi idata idan idapat imemberikan idampak iyang icukup iberpengaruh terhadap ikinerja idan ihasil iclustering. iPada ipenelitian iini, imetode iyang idigunakan dalam iperhitungan ibobot iatribut idata iyaitu iSymmetrical iUncertainty. iUntuk melakukan ipengujian iterhadap imetode iyang idiusulkan, imaka penelitian iini menggunakan idataset idari iUCI iMachine iLearning iyang iterdiri idari iIris idengan jumlah idata isebanyak i150 idata idan iWine iQuality idengan ijumlah idata isebanyak i178 data. iEvaluasi ikinerja clustering yang diusulkan berdasarkan nilai Davies-Bouldin Index (DBI). iHasil ipengujian ipada ipenelitian iini iterlihat ibahwa idengan imetode iyang diusulkan idapat imenghasilkan inilai iDavies-Bouldin iIndex (DBI) iyang isignifikan lebih ikecil.