Abstract
Modern agriculture faces the challenge of increasing production efficiency in the context of limited resources and variable climatic conditions. This article presents an approach to assessing the impact of various factors on agro-industrial indicators using machine learning methods. The primary focus is on the development and application of a hybrid analysis that includes techniques such as gradient boosting (GB), mutual information (MI), and recursive feature elimination (RFE). The study was conducted using data from agro-industrial enterprises in the North Kazakhstan region for the period 2020–2022, encompassing production, climatic, and economic indicators. It was found that crop area, average crop weight, and precipitation are the most significant factors, accounting for up to 93% of the correlation with yield increase. The use of the proposed methods made it possible to reduce forecast uncertainty by 28% and increase the accuracy of key indicator predictions by 15–20%. The results of the analysis, visualized as correlation matrices and feature significance maps, confirm the possibility of applying the proposed approach to optimize the management of agro-industrial production. The application of the developed methodology contributes to the development of strategies aimed at the sustainable development of the agro-industrial complex.