alkan, b. barış sinop üniversitesi - fen-edebiyat fakültesi - istatistik bölümü, SİNOP, turkey , atakan, cemal ankara üniversitesi - fen fakültesi - istatistik bölümü, ANKARA, turkey , alkan, nesrin sinop üniversitesi - fen-edebiyat fakültesi - istatistik bölümü, SİNOP, turkey

A NEW APPROACH FOR ROBUST LINEAR DISCRIMINANT ANALYSIS

28372

Linear Discriminant Analysis is a multivariate statistical method aiming to alocate individuals with known p properties by minimizing the probability of misclassification to real groups (classes) in the natural environment. Linear Discriminant Analysis (LDA) is not a robust method against the presence of outliers in the dataset, and the consequences of using conventional LDA in the presence of outliers in the dataset are far from real. For this reason, it is appropriate to use robust versions of linear discriminant analysis in the presence of outliers. In this study, a new robust version of the LDA is obtained with a combination of the Jackknife resampling approach, minimum covariance determinant (MCD) and LDA method. Proposed new approach is compared with Method-1 proposed by Croux ve Dehon (2001) and Method-2 proposed by Hawkins and McLachlan (1997). These methods will be evaluated through artificial data application and simulation. In the light of the findings the proposed DLDA seemed to give better (or at least as good as them) results in the presence of outliers in the dataset when we compared to the other two approaches based on the classical MCD.

12

Minimumcovariance determinant , robust linear discriminant analysis , outliers

Erciyes University Journal Of The Institute Of Science an‎d Technology

19