Tensor Decomposition Based Visual Analytics Method of Traffic Accident Local Spatio-Temporal Patterns

Traffic accident data contains fine-grained local spatio-temporal patterns.They are difficult to detect because they are often obscured by the full data set.Finding the right subset of data is an important prerequisite for discovering meaningful local spatio-temporal patterns,and it is also the most cumbersome process.For this reason,a visual analysis method of local spatio-temporal patterns of traffic accidents based on tensor decom-position algorithm is proposed to automatically obtain data subsets.The tensor modeling of the traffic accident is performed first,and then the tensor decomposition algorithm is applied to the modeled tensor to obtain a set of vectors which capture the distribution of the dataset in each dimension and can characterize its multidimen-sional patterns.Then,this set of vectors is used to guide the clustering of the data set,thereby dividing the data set into several data subsets that are conducive to discovering local spatiotemporal patterns.A visual analytics system is also designed to support semi-automated exploration of different data subset with local spa-tio-temporal patterns and visual analytics of the obtained patterns.Finally,the visual analysis system is used to explore and analyze the local spatio-temporal patterns of traffic accident data in Hefei.The experimental re-sults are recognized by relevant experts from the Hefei traffic police department,indicating that this method can effectively assist users in discovering the local spatio-temporal patterns contained in the data set.