Discriminant analysis is a statistical method concerned with classifying objects (observations) to previously defined groups. The classical approach discriminant rules are often derived from multivariate normal distribution. The parameters are estimated by MLE (maximum likelihood estimators) method. But, these estimates are highly influenced by outlying observations. The robust estimates are needed. There are two robust estimators. They are robust MCD (minimum covariance determinant) and MWCD (minimum weighted covariance determinant) estimators. The goal of this paper is comparative performance of robust MCD and MWCD estimators in quadratic discriminant analysis. The performance of discrimination functions is measured by the average of proportion of misclassification. The data simulations are generated from various conditions. The variation data consists of the number of groups, the number of outliers, and the kind of outlier: shift, scale, and radial outliers. The performance of robust MCD estimator in quadratic discriminant analysis is the best compared to MWCD and the classical. The average of misclassification proportion of quadratic discriminant function using robust MCD estimator is less than ten percent while the data contaminated by outlying observations are less than 25 percent.

Itulah cuplikan dari abstraksi Jurnal yang berjudul “Perbandingan Kinerja Penaksir Robust MCD dan MWCD dalam Analisis DiskriminanKuadratik.” Melalui blog ini, penulis ingin berbagi pengetahuan mana diantara dua penaksir robust tersebut yang lebih unggul jika diaplikasikan dalam analisis diskriminan kuadratik.