2D-STR: Reducing Spatio-temporal Traffic Datasets by Partitioning and Modelling
Liam Steadman, Nathan Griffiths, Stephen A. Jarvis, Stuart G. McRobbie, Caroline Wallbank · 2019
Spatio-temporal data generated by sensors in the environment, such as traffic data, is widely used in the transportation domain. However, learning from and analysing such data is increasingly problematic as the volume of data grows. Therefore, methods are required to reduce the quantity of data needed for multiple types of subsequent analysis without losing significant information. In this paper, we present the 2-Dimensional Spatio-Temporal Reduction method (2D-STR), which partitions the spatio-temporal matrix of a dataset into regions of similar instances, and reduces each region to a model of its instances. The method is shown to be effective at reducing the volume of a traffic dataset to <5% of its original volume whilst achieving a normalise root mean squared error of <5% when reproducing the original features of the dataset.