The Prediction of Urban Road Traffic Congestion by Using a Deep Stacked Long Short-Term Memory Network

Abstract

Traffic congestion is an overwhelming problem faced by road travelers all over the world. A time-efficient and accurate prediction of upcoming traffic congestion can reduce this problem through enabling the proactive planning of routes. Recent research suggests that prediction accuracy requires the extraction of hidden features of the road network from the historical traffic data. In general, this data is either limited (with a longer sampling time) or not provided by providers. In urban areas, traffic lights, weather conditions, city events, accidents, and people’s habits significantly influence the traffic flow according to the structure of road network. Therefore, a mechanism is required to extract traffic data by scraping images from the route planners’ websites to predict traffic congestion. In this article, we devise such a method and introduce a fuzzy logic and stochastic estimation algorithm to detect congestion levels at the intersections of the road network. We then build a deep stacked long short-term memory network, in combination with online training, for the multipoint future prediction of congestion. We name the proposed model a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">fuzzy logic and deep learning-based traffic congestion predictor</i> (FDLTCP) and compare the proposed predictor with the gated recurrent unit and stacked auto-encoders. Experimental evaluations demonstrate the effectiveness of FDLTCP, in terms of mean square error and other critical performance metrics, to perform future predictions.