Ioulia Markou, who completed her PhD last year in our group, won the 3rd place in the TRA VISIONS 2020 Young Researcher Competition in the cross-modality category, with her work entitled Prediction of traffic anomalies due to special events. Congrats, Ioulia!
Read the description of Ioulia’s work below.
Traffic congestion greatly impact urban areas, as it has significant economic repercussions through deterioration of mobility, safety, and air quality. In 2017, New York drivers spent 91 hours on average sitting in traffic. In Europe, even cities with relatively low congestion levels show excessive network demand during peak hours, which consequently affects the quality of life of their residents. These observations emphasize the need for frameworks that promote better management of city’s road network and transport services. While mature research exists for habitual behaviors, such as commuting cycles, and average situations, current traffic management solutions typically fail under non-recurrent circumstances, such as incidents, special events (e.g. concerts), demonstrations, road works, crisis scenarios (e.g. terrorist attacks), inclement weather, etcetera. This project explores machine-learning architectures for combining time series and textual data for mobility-demand predictions in eventful areas, where abnormal patterns are often observed and not easily explained. Our framework focuses on predicting taxi demand. However, the proposed methodology is applicable to problems that go beyond the transportation domain. Our proposed models significantly reduce forecast errors by using cross-modal sources of information, namely publicly-available taxi and weather data from New York, and information about events collected using web-scrapping techniques and Application Programming Interfaces (APIs). The importance of semantic information is highlighted in all presented methods. The results show that abnormal demand peaks can be accurately detected on time, and competent authorities can subsequently take all the necessary measures to deal with them optimally.