Finding Interesting Regions in Spatial and Spatio-temporal Datasets

When: Monday, April 24, 2017
Where: PGH 563
Time: 11:00 AM – Noon

Speaker: Dr. Christoph Eick, University of Houston

Host: Dr. Shishir Shah

Due to the advances in remote sensors and sensor networks, spatial and spatio-temporal data become increasingly available. In this talk, we present three different approaches to find interesting regions is such datasets. The first approach is a serial, density-based spatial-temporal clustering approach that employs non-parametric density estimation techniques and contouring algorithms to obtain spatial clusters whose scope is described using polygon models; next, it identifies spatio-temporal clusters as continuing polygons in consecutive time frames. The second approach generalizes the popular spatial clustering algorithm SNN to create spatio-temporal clusters. The third approach relies on a graph-based—it employs Gabriel graphs to define object neighborhoods—interestingness hotspot discovery framework which grows hotspots from hotspot seeds, maximizing a plug-in interestingness function. Finally, experimental results obtained by applying the three approaches to air pollution, crime, earthquake and New York taxi cab datasets are presented and evaluated.

Bio:

Christoph F. Eick is an Associate Professor in the Department of Computer Science at the University of Houston and the Director of the UH Data Analysis and Intelligent Systems Lab. His research interests include data sciences, data mining, geographical information systems, artificial intelligence and critical infrastructure resilience. He published more than 150 papers in these areas. He also serves in the program committee of top data mining and artificial intelligence conferences.