Semeniuk R. Methods of processing measuring and expert information using classification scales

The dissertation is devoted to solving an important scientific and practical problem - development of methods for processing experimental data (numerical and verbal) using classification. The paper reviews the scientific works of scientists who worked in the field of classification theory. The main features of classification are defined, among which one of the 20 main ones is the presence of a classification scale developed in advance or in the classification process, based on the condition of distinguishing equivalence classes. In complex classification cases, when the solution is presented in several gradations, the classification quality characteristic is the conditional probability of being assigned to a certain equivalence class. Scientists' experience shows that the statistical phenomena a researcher encounters in the field of classification are complex and not always predictable, therefore methods and techniques for different applications of the classification procedure are very important. This issue is the focus of the author's dissertation, which presents techniques for using classification to work through numerical data, transitioning from numerical data to verbal data using measurement and expert information, and working through verbal data. Among the main points of scientific novelty and practical importance it is worth noting the following: – methods for the non-parametric identification of the shape of the distribution of small samples have been further developed, based on which a method for the non-parametric classification of distribution types has been developed that makes it possible to select an effective estimate of the sample distribution center and thereby reduce the uncertainty of the measurement result; – the method of non-parametric classification of the distribution type was applied for the first time in the construction of process control charts, thus increasing the sensitivity to controllable changes in central tendency and determining the class of the total sampling distribution for process instability; – methods for transition from numerical to verbal data in information and measurement system using a metric classification scale have been further developed, and a theoretical rationale for combining uncertainty in expert information and measurement uncertainty has been developed, that allowed to adequately take into account the uncertainty of the measurement information in the formation of the membership functions of the term set of the scale and to obtain a characteristic of the quality of the established classification scale in the form of a correspondence matrix; – a method for checking the consistency of samples of verbal data obtained behind a metric classification scale has been developed for the first time, allowing consistency checking when equivalence classes are not evenly spaced in their domain of definition; – based on the application of several operators for working with samples of verbal data, the technology for classifying the state of objects based on expert information has been developed, followed by the use of the resulting classification scale to build process control charts for qualitative attributes. Theoretical and practical results of the thesis are connected with the scientific themes and plans of the department of information and measuring technologies of Igor Sikorsky Kyiv Polytechnic Institute, as well as research paper of the state budget / fiscal theme «Research and development of metrological and technical support for the implementation of environmental monitoring tasks in Ukraine» (№ RP 0113U002159) and research paper «Theoretical and practical principles of the application of soft measurements in product quality systems» (№ RP 0118U001478).