Date of Publication :31st May 2023

Abstract: A problem statement like "traffic congestion" has a wide variety of implications on society and the economy. Work is continuously being done in this area to make significant advancements. We have attempted to anticipate the network-wide traffic flow speed using time series analysis and cutting-edge deep learning techniques. For our work, we took into account the historical traffic data for Chicago, which includes the speed of the next time period's 1047 individual road segments. We converted the traffic data into a spatio- temporal matrix and added temporal data for each spatial road segment separately in each column because the traffic data contains time series for each spatial road segment. A RNN model with two layers of LSTM that allots one memory unit to each road segment was created. Using the spatio- temporal training matrix, we trained our model for 50 epochs and then received a vector containing the speeds of each road segment for the subsequent time step. For both the training set and the validation set, the model clearly displayed a learning tendency. Finally, for better visualization, we calculated the MSE and RMSE for the model on the spatio- temporal test matrix and also rendered the prediction as a spatio-temporal image.

Reference :