Mining Unusual Patterns by Multi-Dimensional Analysis of Data Streams

It has been popularly recognized that stream data represents an important form of data, with broad applications. There have been a lot of studies on effective stream data management and query processing, as well as some recent studies on stream data mining. Although this is a promising direction, most existing studies have not paid enough attention to one critical fact: most data streams reside at a rather low level of abstraction and are multi-dimensional in nature, whereas most analysts are interested in finding characteristic features, unusual patterns, and dynamic changes (such as trends and outliers) at relatively high levels of abstraction and in certain multi-dimensional space. To accomplish such tasks, one may need to develop effective mechanisms for on-line, multi-dimensional analysis and mining of stream data. This poses great challenges on system architecture, implementation methodology, algorithm development, and performance tuning. In this paper, we discuss the issues related to effective, on-line, multi-dimensional analysis and mining of unusual events and patterns in data streams, including research challenges, potential architectures, and implementation methodologies.