Zabielin S. Models and methods of forecasting volcanic activity using artificial intelligence technology

Zabielin S. Models and methods of forecasting volcanic activity using artificial intelligence technology. - Qualifying scientific work, the manuscript. Thesis for a PhD degree in specialty 122 "Computer Science". – National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute", Kyiv, 2020. In the dissertation the following new scientific results were first received: A method of variable selection was developed for the first time, characterized in that it uses the results of the forecasting system to iteratively improve the selected variables. Improved iterative method of reducing the dimension based on the method of Isomap, which differs in that it uses feedback from the test system, which allows to achieve high informativeness of variables. Improved multi-criteria selection task, characterized in that the selection of variables uses levels of confidence, expert assessment of significance, to improve the accuracy of the model. A new genetic algorithm has been developed to generate alternatives in the formation of a forecast that differs from the existing ones, using a special polynomial that connects all input variables and allows to reveal hidden relationships within variables. A new method of variable selection has been developed that differs from the existing ability to efficiently process large amounts of input variable, due to the presence of several independent criterions of the selection. Keywords: neural networks, genetic algorithm, variable selection, dimensionality reduction, fuzzy numbers, volcanoes, volcanic activity, forecasting.