In the context of globalization, international scientific migration is becoming critical for the sustainable development of national innovation systems, acting as a key factor in the dynamics of human capital. This study proposes a comprehensive approach to analyzing and forecasting migration flows of scientists based on the integration of system dynamics methodology and machine learning methods. A system-dynamic model has been developed and validated that represents the scientific labor market as a polysystemic formation and analyzes local patterns in weighted directed temporal migration networks. Highly accurate models of short-term forecasting of the relative migration balance of scientists have been built using deep neural network stacking and ensemble methods. The developed model complex has proven its effectiveness as a tool for supporting strategic decision-making in scientific, technical and migration policies aimed at preserving and developing scientific human capital.