Controlling the number of regressors in linear equations when constructing a listwise regression model

Download article in PDF format

Authors: Bazilevskiy M. P.

Annotation: For clustering an available sample of statistical data, a listwise regression model is proposed that contains a complete set of input variables in each equation of the list. The problem of esti-mating unknown parameters of this model using the least abso-lute deviations method is reduced to a mixed 0-1 integer linear programming problem. To control the number of regressors in the list equations, the optimization problem is extended with additional constraints. Solving this problem makes it possible to obtain the best subset of regressors included in the list equa-tions, the equations’ coefficients, and the switching rule be-tween them. Computational experiments were carried out to confirm the correctness of the developed mathematical ap-proach.