Approach to Solving the Multicolinearity Problem Using Quasi-Orthogonalization of Empirical Data

Abstract

This article proposes an approach to solving the problem of regressors multicollinearity using the procedure of quasi-orthogonalization of data. The specified approach is based on the transformation of factors during their coding according to the rules of a full factorial experiment. It is shown that the proposed coding of factors leads to a reduction of multicollinearity of the data. This approach can be used both for building models based on short samples and for batch processing of Big Data.