Approaches to Identification of Linear Relations from Compound Noisy and Noise-Free Data


Dragan Radojević




Frisch scheme defines the whole set of possible solutions of linear relations for numerical series in general case. This paper presents a modification to the Frisch scheme made for the following special case: some series of numbers are noise- less—independent variables; and the remaining are noisy — dependent variables. Special results from the set defined by modified Frisch scheme, which represent potential solution for prediction are analyzed. The result obtained on unweighted principal components of the matrix of covariance of explained part of variance of dependent variables, represent the best solution for prediction.