City-Level Traffic Flow Prediction via LSTM Networks

Abstract

Along with the rapid development of modern urban cities and the increasing number of vehicles, they also cause severe traffic problems. They trouble citizens a lot in wasting time and gas energy, leading to their bad temper during work, and even causing traffic accidents. Thus, monitoring the traffic flow is a critical portion in building intelligent transportation systems for urban cities. In this paper, we aim to predict traffic flow with big traffic data in urban cities. Our intuition is that although the traffic information is influenced by many factors and precise traffic modeling is difficult, we use the machine learning techniques to analyze the historical traffic data and train the prediction model. Particularly, we build a LSTM model with different dimensions of historical data, and then make prediction almost in real time for current traffic flows. We have built a prototype and tested it with two benchmark dataset. Experiment results show that our system obtains high prediction accuracy compared with the ground truth.