As the world becomes instrumented and interconnected, spatio-temporal data are more ubiquitous and richer than ever before. Moving object (e.g., taxi, bird) trajectories recorded by GPS devices, social event (e.g., microblogs, crime) with location tag and time stamps, and environment monitoring (e.g., remote sensing images for weather forecasting) are typical spatio-temporal data that we meet every day. These emerging spatio-temporal data also bring new challenges and opportunities to data analytics research and business intelligence solution: Can we discover more predictive patterns from combining space and time dimensions, and better solve real-world business problems?
The first law of geography tells us that “everything is related to everything else but nearby things are more related than distant things”. Such a characteristic is also known as the spatial autocorrelation. Therefore, the widely used i.i.d. assumption in data mining is too strong when analyzing spatial data. New methods and modeling techniques are needed to tackle with the spatial heterogeneity and the spatial relationships (such as topological relationships, directional relationships, etc.), which are unique to spatial data. Spatio-temporal data are further temporally dynamic, which requires explicit or implicit modeling the spatio-temporal autocorrelation and constraints to achieve good prediction performance.
In real world, we also face great challenges from massive data volume, data uncertainty, complex relationship, and system dynamics. Thus spatio-temporal analytics research will focus on the following topics:
Some of the components have been developed and contributed to IBM predictive analytics software such as SPSS Modeler, and industrial solutions such as Crime Information Warehouse (CIW) and asset failure pattern analysis.
Typical business scenarios of spatio-temporal analytics include: