Anomaly detection in taxi flow by a projection method

Abstract

The prevalence of location-aware devices has propelled studies on the analysis of movement data. In this study, we investigated anomaly detection in taxi trajectories. A projection method was used to detect trajectory outliers. This is a robust statistical method and considers two dimensions of data simultaneously—distance and time. The experimental data included taxi movements in Seoul City and New York City. The results were compared with those of an alternative method, the Mahalanobis distance approach. The findings were observed to be similar. The proposed method can be used to improve the services of taxis and buses.