Subject of research: mixed-integer 0-1 linear programming problem for choosing optimal structures for non-elementary linear regression models.
 Purpose of research: integrate into the mixed-integer 0-1 linear programming problem additional constraints that will guarantee the construction of quite interpretable non-elementary linear regressions.
 Methods of research: regression analysis, mathematical programming, method of successive increase the absolute contributions of variables to the general determination.
 Object of research: non-elementary linear regression models.
 Main results of research: in the mixed-integer 0-1 linear programming problem, designed to construct non-elementary regressions, additional linear constraints on the absolute contributions of variables to the general determination are integrated, allowing you to control both the contributions themselves and multicollinearity in the model. It is shown how it is necessary to regulate these constraints so that the non-elementary linear regression obtained as a result of solving the problem is quite interpretable. The proposed mathematical apparatus was used to model railroad freight transportation in the Tyumen region. An interpretation of the obtained high-precision and quite interpretable non-elementary linear regression is given.