Event-driven anomalies in spatiotemporal taxi passseger demand

We propose a grid-based approach of mining historical taxi GPS data to study anomalous pickup patterns in spatiotemporal passenger demand caused by public events. The developed method is applied by way of example to the city of Munich on a large-scale taxi dataset. Information about occurring public events is collected from different online and offline sources. For further analysis, the data is aggregated on an equally distributed grid level based on geohashes. In a first step, an extension of robust principle component analysis (RPCA) is developed and applied to the dataset. It is shown that extended RPCA can separate anomalous pickup patterns from base demand in the given time series. Secondly, the combination of anomaly detection and knowledge about occurring public events allows the exploration of correlations in demand fluctuations. The results show that the proposed method can be used to discover the potential impact of public events on spatiotemporal passenger demand in large-scale mobility datasets.