Time series, prediction, ARIMA.

This paper introduces a development method for time series prediction based on averaging values. The experimental part will focus on teaching more neural networks with the same topology and settings that will solve the same problem (time series). The resulting values for training and test set are averaged depending on how many neural networks are involved in the calculation. The experimental part is focused on testing of periodic time series with different topologies and neural network settings. The results of prediction are compared with ARIMA models.