Short-term real-time traffic prediction methods: a survey

Abstract

Short-term traffic prediction provides tools for improved road management by allowing the reduction of delays, incidents and other unexpected events. Different real-time approaches provide traffic managers with varying but valuable information. This paper reviews the literature regarding model-driven and data-driven approaches focusing on short-term real-time traffic prediction. We start by analyzing real-time traffic data collection, referring network state acquisition and description methods which are used as input to predictive algorithms. According to the input variables available, we describe common and useful traffic prediction outputs that should contribute to understand the panorama verified on a road network. We then discuss metrics commonly used to assess prediction accuracy, in order to understand a standardized way to compare the different approaches. We list, detail and compare existing model-driven and data-driven approaches that provide short-term real-time traffic predictions. This research leads to an understanding of the many advantages, disadvantages and trade-offs of the approaches studied and provides useful insights for future development. Despite the predominance of model-driven solutions for the last years, data-driven approaches also present good results suitable for Traffic Management usage.