Instruments and Systems: Monitoring, Control, and Diagnostics Annotation << Back Automatic Generation of Pipelines for Event Detection in Proactive Monitoring Systems of Complex Systems M.A. BOLSHAKOV, A.D. KHOMONENKO, V.A. KHODAKOVSKY Automatic generation of coherent chains of models, called pipelines, allows to realize different logic of data preparation for the fi nal learning algorithm, thereby increasing the quality of the fi nal model. However, the large variety of algorithms and the generally unknown nature of their interaction does not allow for expert selection of the best combination of methods. For this purpose, either brute-force methods are most often used, which in most cases is computationally expensive, or automatic machine learning methods. There are not aimed at optimizing the whole payplane, or solve the problem in the most general form, without taking into account the specifi cs of the problem to be solved. And this also eventually leads to a large computational cost for the selection of models and their parameters. The paper proposes to take into account the peculiarities of monitoring tasks, primarily related to the representation of analyzed data in the form of multidimensional time series with rare events-incidents, which it is desirable to identify with some advance, which will narrow the search area and optimize computational costs. Keywords: monitoring, machine learning, pipeline, condition assessment, dimensionality reduction, sampling. DOI: 10.25791/pribor.12.2024.1547 Pp. 48-59.