INFLUENCING FACTORS INFORMATION USAGE FOR SPLITTING DATA SAMPLES IN MACHINE LEARNING METHODS TO ASSESS IS STATE

Improving the qualitative indicators of identifying the state of information security of individual
cyber-physical systems segments is associated with the processing of large information
arrays. A method of splitting data samples is proposed to improve the quality of algorithms for classifying
information security states. Classification models are configured on training sets of examples
in which outliers, noisy data, and an imbalance of observed objects may be present, which affects the qualitative indicators of the results. At certain points in time, under the influence of the external environment, the frequency of occurrence of observed events, the ranges of recorded values may
change, which significantly affects the quality indicators. It is shown that a number of events in the
samples occur as a result of the actions of internal and external factors.