ENHANCING LARGE SCALE TRAFFIC CONGESTION PREDICTION WITH ATTENTION-AUGMENTED BILSTM NEURAL NETWORK

Abstract

One major issue impeding the sustainable growth of urban traffic is traffic congestion. To avoid traffic congestion, it is crucial to assess the current state of the traffic and project future traffic patterns. Identifying and creating prediction techniques for a city wide in urban setting is the key objective of this study. Predicting long-term levels of road congestion can help commuters avoid crowded regions and allow traffic agencies to take the necessary measures. In this paper, we present attention-augmented BiLSTM neural network approach for hourly based monthly traffic congestion prediction. Our experiments show that the proposed model outperforms the baselines in terms of accuracy and error.An increasing number of studies are using real -time data-which is gathered using various devices including GPS, loop detectors and fixed location traffic sensors-to improve the prediction impact. Fixed location traffic sensors are more affordable, therefore, in our research we estimate the traffic congestion using real-time data obtained by these sensors.