(ICCC 2020) Traffic Incident Detection from Massive Multivariate Time-Series Data

Abstract: Smart-city infrastructure has the potential to improve the lives of anyone who finds themselves in an urbant environment. However, Vehicle Tranffic Man- agement (VTM) which ameleorates traffic congestion to improve productivity and reduce commuter stress is hampered by traffic incidents which introduce unexpected and chaotic variability into the traffic network. Automatic Incident Detection (AID) mechanisms aim to quickly and reliably identify vehicle traffic incidents to reduce the effect of traffic incidents on VTM. This paper shows that widely available magnetic traffic sensor data can be used with the AdaBoost Ma- chine Learning (ML) model to produce a reliable, light-weight AID mechanism to assist with VTM. A comparison with other ML models is also presented as well as ideas for future work.

Authors: Nicholas A Sterling and John Miller (University of Georgia, USA)

Email: nickbk@uga.edu, jam@cs.uga.edu