A neural network algorithm for the prediction of events in multidimensional time series and its application to the analysis of data in cosmic physics

Many practical problems are related to the search for interconnections between the behavior of complex objects and relatively rare events caused by this behavior or correlated with it. In such cases, it can be assumed that the occurrence of each event is preceded by some phenomenon, i.e., a combination of values of the features describing the object under consideration in a known range of time delays. This work continues the investigation of the neural-network based method for analyzing such objects developed by authors elsewhere. The method aims at revealing morphological and dynamical features that cause the event or precede its occurrence.[PUBLICATION ABSTRACT]