The approach to the preliminary analysis of the multidimensional time series, applied to develop the intellectual classifiers in the problems of proactive maintenance of the complex systems, has been studied. It is proposed to execute the data analysis with application of the UMAP algorithm, which implements non-linear dimension reduction of data. It is made for the purpose of the data visualization and early identifying of the border states on the boundary of the classes, which separates normal and abnormal states of the complex systems, confirmed by the facts. The examples of the preliminary analysis of the theme dataset are given. This dataset is represented by the multidimensional time series, which characterize normal and abnormal operation of the aero engines in various time moments. These time series were obtained with application of the appropriate sensors. The possibility of the early prediction of the abnormal states (damages) of the aero engines, identified after their emergence, was confirmed.