Pavlov A. Technology for building regression models based on an iterative algorithm with recurrent computations

The thesis is devoted to development of a technology for building regression models based on a generalized relaxational iterative algorithm GRIA GMDH with recurrent computations for solving high-dimensional applied modeling problems. Nonrecurrent and recurrent methods for parameters estimation and selection criteria calculation have been newly developed for relaxational iterative algorithms. A new numerical investigation method for iterative algorithms convergence rate has been developed. The method allows increasing statistical reliability of algorithms efficiency estimation. A new information technology and software for solving high-dimensional modeling problems on the basis of experimental data have been developed and implemented. Solution results for real-world applied modeling problems are given: space weather forecasting (the model forecasts every two of tree magnetic storms); medicines efficiency estimation (the model accuracy is 70%); differential diagnostics of mild forms of hemostasis pathologies (recognition accuracy is 89%)