engleski

Clustering of the Anomalous Spatiotemporal Traffic Patterns Using Tensor Decomposition Method

Tensor-based models emerged only recently in the field of traffic data analysis. They outperform other data models because they can simultaneously capture both spatial and temporal components of the observed traffic data. In this paper, the Non-negative Tensor Decomposition (NTD) the method is used to extract traffic patterns in the form of Speed Transition Matrices (STM). The traffic pattern anomaly is estimated using Kullback–Leibler Divergence (KLD) between the observed traffic patterns and the average traffic pattern. Anomalous traffic patterns are then clustered using Agglomerative Clustering (AC), regarding its temporal components. Experiments were conducted on the large sparse Floating Car Data (FCD) for the most relevant road segments in the City of Zagreb, Croatia. Results show that the method was able to detect and cluster the most anomalous spatiotemporal traffic patterns, representing the traffic on the observed road segments. Valuable traffic insights can be extracted by using the proposed method, including the location and the cause of traffic anomalies. Therefore, such traffic information can be used in routing applications and urban traffic planning.

clustering ; non-negative tensor decomposition ; traffic anomaly ; sparse gnss data ; speed transition matrix

nije evidentirano