The paper investigates the problem of planning production processes at the stage of raw material allocation. For a given set of requirements (production orders), the execution volumes and acceptable limits for various impurities are specified. Similarly, the volumes of each raw material resource are given. However, the levels of impurities in raw material composition are determined only approximately, which significantly complicates the planning process due to obvious risks associated with achieving solution quality indicators. To solve this problem, a mixed-integer linear programming model (MILP) is proposed as the primary planning tool. This model also implements an approach to form maximal feasible subsystems of constraints based on the penalty function method. At the stage of creating the functional space for the model, we use regression analysis methods, such as linear regression, random forest, and gradient boosting, to predict the qualitative characteristics of resources. A computational experiment is conducted on both simulated and real production data from an industrial enterprise in the non-ferrous metals industry. Another direction of the computational experiment is the comparative analysis of the efficiency of open-source libraries PuLP and HiGHS in Python, for solving optimization problems within the framework of the proposed MILP model.