The global transformation of intellectual labor markets actualizes the development of precision tools for managing mobile scientific human capital. A hybrid structural-measurement model for forecasting migration flows of scientists based on machine learning methods is proposed. The model is developed using panel data from different countries. The MICE method with protection against temporary data leakage is used to handle missing values. Comparative analysis of algorithms (Random Forest, XGBoost, ANN, Stacking) revealed the superiority of ensemble methods, with the best result for Random Forest. SHAP analysis identified the key determinants of migration. The results are tested using cross-validation and stability analysis. A statistically significant relationship is established between migration intentions and economic factors. Institutional risks are identified, including the phenomenon of excess qualification, leading to an underestimation of the forecast of the outflow of intellectual capital. Practical mechanisms for optimizing migration policy are proposed: an early warning system based on ML models; infrastructure bonds for financing innovative clusters; differentiation of academic mobility and permanent migration. The study contributes to the theory of human capital, offering formalized approaches to managing intellectual migration in the context of global transformation of labor markets.