Abstract

In the present paper, several test samples with scattered data and outliers were examined by means of different methods for building linear regression equations. This includes ordinary least squares, least absolute deviation, orthogonal distance regression, and the least absolute deviation of orthogonal distance. Also, a variant of the weighted least squares approach was investigated in the problem of identification of outliers. New indices have been proposed to describe the outliers, which can be viewed as weighted coefficients of determination.

Keywords: linear regression, least squares, determination coefficient, outliers

Article language: En.