Comparative analysis of methods of best regression structure search. Volodymyr Stepashko,Iana Bondarska

Abstract. Most critical part of linear regression algorithms is selection of structures to be checked. One way to reduce the number of such structures is to make some primary estimations of structures quality and to divide the structures into groups according to this estimation. One technique that uses this approach is La Motte-Hocking algorithm. This paper represents research results of La Motte-Hocking method efficiency in comparison with the combinatorial algorithm. Results of research and some possible ways to improve results are presented in this paper. The outcomes are planned to be used for updating the combinatorial GMDH algorithm.

Keywords. Linear regression, best subset search methods, La Motte-Hocking method.

References.

1. Statistical Methods for Digital Computers. Edited by Kurt Enslein, Anthony Ralston and HERBERT S. Wilf. John Wiley, New York, 1977