This study is devoted to researching the problem of traffic congestion related to the growth of the number of vehicles in the world, in which the analysis of experiments to predict traffic flows using deep learning models is presented. Recurrent neural networks including the Elman model, LSTM, and GRU model were selected for the experiments, and the MSE indicator was used to compare their performance. The results showed that the GRU model performs best in predicting traffic flows, making it a promising tool for traffic management and infrastructure optimization.