[en] Machine learning is one of the popular methods for analyzing and processing complex data. Despite it shown good accuracy, applying it in the scientific field is hindered by the unpredictable neural networks behavior. Thus, incorrect results can be caused by applying neural networks to separate particles in scintillator. Therefore, it was necessary to compare series of different neural networks architectures and to find out the feasibility of their application to the task of separating particles according to the shape of the pulse.