| Summary: | Background. Constantly growing requirements for ensuring the safety and reliability of technical
systems lead to the need for more accurate diagnostics of the state of the facility in operating conditions based on the
results of monitoring the performance of this facility. Sometimes it is necessary to describe the state of an object using
several possible options. In this case, a multiclass classification is carried out, in which the possible states of the object
are divided into several classes, for example, by the type of failure. At the same time, machine learning methods can
be effectively used. Features of the problem under consideration are a limited amount of sample data, as well as the
imbalance of the training sample: information on the performance indicators of functioning in inoperable states of the
object, as a rule, is much less than in the case of able-bodied ones. The purpose of the study is to develop a technology
for diagnosing the state of a technical object according to the specified indicators of its functioning, taking into account
these features. Materials and methods. Among the machine learning methods used for multiclass classification are both standard statistical and special: neural networks, compositional models, aggregated classifiers. In this paper,
the Random Forest method was used for multiclass classification , which showed high quality in solving various machine
learning problems. Results and conclusions. A technology for multiclass diagnostics of technical systems using
a random forest in the Statistica system has been developed. On the example of diagnostics of a computer system, it is
shown that the use of this method provides a sufficiently high accuracy of classification. In case of class imbalance, F
is used as a classification criterion instead of the error rate. – Measure. If necessary, the number of performance indicators
can be reduced taking into account their importance.
|
|---|