The paper considers the issues of increasing the effectiveness of surrogate modeling based on sequential optimization of data sampling. The current state of research in the field of developing algorithms for clarifying the experimental plan for surrogate modeling is analyzed, shortcomings and directions of development are identified. The use of genetic algorithms in combination with machine learning methods is proposed to meet the requirements for efficiency and speed of solving complex engineering problems. A genetic algorithm for sequential optimization of data sampling to build surrogate models has been developed, based on the basic sampling algorithms of Surrogate Modelling Toolbox library and having various modifications, including different selection strategies, mutation and crossing methods. The use of an evolutionary approach to find optimal points in the parameter space of machine learning models based on the estimation of the largest error of the surrogate according to the plan of the computational experiment is substantiated. The implementation of the approach for various sampling algorithms is presented: Random Sampling, Full-factorial Sampling, Latin Hypercube Sampling. The effectiveness of the developed algorithm to solve the regression problem on the Ackley’s and Rastrigin’s test optimization functions is evaluated based on numerical experiments.