Outlier Detection in Urban Traffic Data

This paper provides a summary of the tutorial on outlier detection in urban traffic data. We present existing solutions in three main categories: Statistical techniques, similarity-based techniques, and techniques based on pattern analysis. The first category groups solutions employing statistical models to identify anomalies in traffic data. The second category groups solutions using distance measures and neighborhoods to derive local density estimates. The third category explores the correlation between traffic flow values by using concepts from pattern analysis. We explain and discuss example solutions for each category, and we outline perspectives on open questions and research challenges. We relate the solutions to a general view on the notion of locality, i.e., the context and reference used in the definition and comparison of outlierness, in order to gain a better understanding of the intuition, limitations, and benefits for the various outlier detection methods for urban traffic. This way, we hope to provide some guidance to practitioners for selecting the most suitable methods for their case.