A Comparative Analysis on Traffic Flow Prediction

Over the last few decades, as the population has been increasing day-by-day, similarly there has been an exponential increment of vehicles across the world. Due to which achieving the accuracy in the traffic flow prediction is crucial for solving the factors which are affecting the traffic flow, such as: traffic congestion, road accidents, special events, road closure, etc. So, to counter the aforementioned issues, several number of traffic flow prediction technologies have been developed in recent years, the majority of which rely on short-term traffic prediction, which includes parametric and non-parametric models. We have explored several traffic predictions models in our paper. In this study, the efficiency of six representative prediction models is compared by taking historical traffic data of San Francisco with 10-minute time intervals. These are multiple linear regression, polynomial regression with degrees two and three, LSTM, bidirectional LSTM, and stacked LSTM. The result demonstrates that bidirectional LSTM outperforms the remaining models significantly. We believe that, those individuals who desire to learn more may find this article useful.