Abstract:
A method for generating synthetic data for training systems in binary Markov data sources is presented, based on estimates of the elements of the transition probability matrices of binary Markov chains obtained as a result of a multiscale analysis, which differs from the known ones by taking into account the ranges of values of the matrix elements in the observed objects. An algorithm for the formation of synthetic data is proposed, which implements the calculation of elements of transition probability matrices within the estimates obtained on real data. The results of a computational experiment organized to test the quality of machine learning using the developed method and algorithm confirmed the possibility of improving the quality of artificial intelligence systems.
