Basing a decision on a single data model can lead to prediction impairment. Although a training dataset is considered as an epitome of out-of-sample entries, including external dataset will play a nuisance role by adding more confounding variables to the original endogenous variables. In this project, we present a cross-exogenous multi-modeling technique for blend probabilities for predicting traffic flow. Encouraging results were achieved by stacking the Restricted Boltzmann Machine combined with a multitask prediction layer, the latter being fully connected perceptron layer or Gated Recurrent Unit. Promising results could be achieved by using a backward rolling window to calculate the auto-correlation to find an optimized window size.
After examining results and publication of refereed papers of research in this field, Theoretical research will address the scientific challenge and will study the approach of providing multiple good solutions by using smart technologies, which can be used for Traffic Flow Prediction for traffic infrastructure. Moreover, by establishing this state of the art in the first phase of the projects, members want to make sure that actual traffic data sets can be used and shared among them as soon as possible to test software library implementations.
When employing tools of artificial intelligence in Traffic Flow Prediction, the utility of our software will help infrastructure to improve the quality of transportation. Therefore, the researchers intend to create a software system that simulates smart Traffic Flow Prediction with database mining automated tools for estimating Deep learning Neural Networks.