Anomaly Detection in Vehicle Traffic Data Using Batch and Stream Supervised Learning

Abstract

Anomaly detection in the execution time of vehicular traffic routes is important in smart cities context. This is due to identified anomaly situations allow replanning traffic in large cities. However, this is a problem of difficult treatment, since it is difficult to label such data. This fact leads to some works in literature tackle the problem using unsupervised or semi-supervised learning algorithms. Techniques for labeling data allow using supervised learning. On the other hand, data streams raises questioning the performance of batch and stream learning algorithms. The purpose of this work is to present a process of anomaly detection in the execution time of vehicle traffic routes. Part of the proposed process was evaluated, using data collected in real scenarios. The results were promising for two of the evaluated routes.