Enhanced mobility awareness : a data-driven approach to analyze traffic under planned special event scenarios

Abstract

Traffic disruptions impose societal costs of billions of dollars every year. A constant increase in mobility demand, combined with ongoing urbanization, exacerbates the problem. Since extensions of the infrastructure are for the most part no longer feasible, researchers are trying to find solutions to increase the efficiency of the road network usage. One key element to meeting that goal is to use smart prediction techniques on as many traffic-influencing factors as possible. With the availability of traffic datasets with high spatial and temporal resolutions, more and more data-driven solutions to predict the impact of these factors have been presented by the community. However, while the impacts of hazards, road accidents, and daily rush hour have been the subjects of intense study and analysis the specific impact of so-called planned special events on traffic remains mostly unexplored. Are the effects of upcoming concerts, sporting events, etc. predictable at all? This is the main question that we address in this thesis. We focus our analysis on three different aspects. First, we analyze the general characteristics of event-caused traffic disruptions around different venues in Germany. The results show, that the impact of events varies strongly, being highly affected by its venue location, the time of day, and the event category. In the second step, we analyze the spatial impact of events around different venues. This spatial impact describes a set of road segments, that people tend to use to get to and from the venue. To identify those preferred routes, we propose a classification-based technique that measures event influence for each road segment separately. The approach is based on a large scale analysis across many different venues in Germany. Results show impact zones around several soccer venues in Germany that we discuss in detail. In the third part of this thesis we analyze features from online sources (Twitter, Facebook, etc.) in terms of their explanatory power towards the expected event impact. We collect a large list of different information sources for major events in different venues. Based on that collection, we present prediction models for various measures of event impact. Our results show, that these approaches are capable to predict the severity of event impact under certain conditions, which allows decision makers to create traffic management strategies tailored to event caused traffic disruptions.