Application of Sampling Algorithms for Sample Balancing in Diagnostics of Technical Object Condition

The paper proposes the use of sampling algorithms for the purpose of balancing the sample when diagnosing the condition of technical objects. The efficacy of a number of methods, including SMOTE, Tomek links, and ADASYN, is considered, as well as hybrid algorithms: TaS - a combination of Tomek links and SMOTE, and proposed by the authors new algorithm TaA - a combination of Tomek links and ADASYN. The support vector method is employed as the training model. The predictive quality of the models is evaluated using the F- score and A UC ROC metrics. In order to prevent the model from being fitted to achieve optimal predictive performance, cross-validation is conducted. The proposed methods are implemented as a software program written in the Python programming language. An analysis of the effectiveness of various sampling algorithms for sample balancing with the hydraulic system valve serving as a case study has been conducted. Three scenarios with disparate ratios of class imbalance, that is, the proportion of faulty conditions relative to serviceable ones, are examined. It is shown that the use of sampling methods enables balancing the sample and enhancing the values of quality metrics in predicting by approximately 30-40% in comparison to models that do not incorporate data balancing. This, in turn, facilitates the timely detection and warning of potential facility failures. The findings of the study indicate that hybrid data balancing techniques, such as TaS and TaA, are more efficacious than standalone methods, specifically, SMOTE, Tomek links, and ADASYN.