PERBANDINGAN METODE KEKAR BIWEIGHT MIDCOVARIANCE DAN MINIMUM COVARIANCE DETERMINANT DALAM ANALISIS KORELASI KANONIK

Riana, Freza
Hamim Wigena, Aji
., Erfiani

Krea-TIF: Jurnal Teknik Informatika
Vol 3 No 2 (2015)

Canonical Correlation Analysis(CCA) is a multivariate linear used toidentify and quantify associationsbetween two sets of random variables. Itsstandard computation is based on samplecovariance matrices, which are howeververy sensitive to outlying observations.The robust methods are needed. Thereare two robust methods, i.e robustBiweight Midcovariance (BICOV) andMinimum Covariance Determinant(MCD) methods. The objective of thisresearch is to compare the performanceof both methods based on mean squareerror. The data simulations aregenerated from various conditions. Thevariation data consists of the proportionof outliers, and the kind of outliers: shift,scale, and radial outlier. Theperformance of robust BICOV method inCCA is the best compared to MCD andClassic