%0 Journal Article %A Fainzilberg, Leonid S. %D 2020 %I Begell House %K Discriminant function, diagnostic attributes, self-organization methods %N 5 %P 38-50 %R 10.1615/JAutomatInfScien.v52.i5.40 %T Plausible but Groundless Premises when Constructing Diagnostic Models %U https://www.dl.begellhouse.com/journals/2b6239406278e43e,498cb5837c4d86a8,35cce7ab35991930.html %V 52 %X When solving a number of applied problems of medical and technical diagnostics, the designing of diagnostic models is carried out under the conditions of insufficient knowledge of physical regularities occurring in the studied objects. It is necessary to create models only based on "common case" and intuition, relying on the available experimental material (precedents). However, here erroneous solutions are possible that lead to the inefficiency of the diagnostic system. We consider examples of some considerations, which turn to be scientifically insolvent. It is shown that the linear classifier, to which the method of decision-making on the distance to standards is reduced according to the method of reference images, can lead to absurd results. Such an effect occurs if the independence condition for the preferences of individual characteristics is not fulfilled, which foresees the "worsening" of the value of one attribute can be compensated by the "improvement" of another, and vice versa, which is not always true. It is shown that the unjustified expansion of the space of diagnostic attributes impairs the effectiveness of the diagnostic rule. Therefore, it is important to get rid of unnecessary signs even before the training stage. We analyzed the inconsistency of the argumentation that when constructing diagnostic models it is expedient to use only statistically independent attributes. For showing the fallibility of such an argumentation it is proved that under a statistical dependence between attributes, a combination of individually non-informative attributes can be not only useful but also capable of providing unmistakable recognition of classes. Therefore, it is important in every definite case to investigate the question of conditional statistical dependence between attributes before making decision on their exclusion from a description. By the example of constructing a model of indirect estimation of the carbon content in a liquid metal by the temperature of crystallization onset, it is shown that it is impossible to restore the true diagnostic model using only self-organization methods without using additional algorithms. %8 2020-09-15