Research Abstracts Online
January 2010 - March 2011
University of Minnesota Twin Cities
College of Science and Engineering
of Computer Science and Engineering
PI: Shashi Shekhar, Fellow
Multi-Resolution Spatio-Temporal Cascade Pattern Mining
Multi-scale spatial data mining (MSDM) aims to identify interesting, important, and non-trivial spatial patterns that show up at multiple analysis scales. An important instance of the MSDM is the multi-scale spatial outlier detection problem. A multi-scale spatial outlier (MS-Outlier) is a location that is significantly different from its spatial neighbors at some analysis scales. Given a set of spatial neighborhood scales, the problem aims to find all such kind of MS-Outliers with their outlying scores at each scale. The MS-Outlier detection problem is important for applications like mapping and decision making for public safety (e.g., crime analysis), and climate science (local, regional, and global climate phenomenon discovery).
The MS-Outlier detection problem is challenging due to the requirements of both finding a generic method to model dynamic spatial relationships and pattern interestingness at various scales and efficiently computing outlier scores at multiple scales. These researchers are developing a novel representation schema for the multi-scale spatial data-mining problem. They are designing efficient algorithms to detect MS-Outliers and evaluate the computational performance of various design decisions using a real precipitation dataset.
Michael Robert Evans, Graduate Student
Visanath Gunturi, Graduate Student
Zhe Jiang, Graduate Student
James M. Kang, Graduate Student
Pradeep Mohan, Graduate Student
Dev Oliver, Graduate Student
Zhou Xun, Graduate Student