Traffic Flow Prediction Using Deep Learning Models

Abstract

For the effective implementation of Intelligent Transportation Systems (ITS), accurate and timely traffic flow information is critical. Traffic data has exploded in recent years, ushering in the era of big data. The main issue of a traffic flow prediction system is determining how to build an adaptive model based on past data. Existing traffic flow forecast approaches rely on shallow learning models, which are unsatisfactory for many real-world applications. This situation prompts us to reconsider traffic flow prediction using deep learning models and large amounts of traffic data. In this paper, we present the Long Short-Term Memory (LSTM) model, Gated Recurrent Unit (GRU) model, Stacked Auto Encoder (SAE), and an Auto-Regressive Integrated Moving Average (ARIMA) model, all of which are deep learning methods. These techniques are applied to real-world traffic big data collected by performance measurement systems. Compared to other deep learning and shallow machine learning prediction networks, experimental results demonstrate that the gated recurrent unit model is more applicable and performs better.