MINING TEMPORAL SEQUENCES TO DISCOVER INTERESTING PATTERNS
Edwin O. Heierman, G. Michael Youngblood, Diane J. Cook · 2004
When mining temporal sequences, knowledge discovery techniques can be applied that discover interesting patterns of interactions. Existing approaches use frequency, and sometimes length, as measurements for interestingness. Because these are temporal sequences, additional characteristics, such as periodicity, may also be interesting. We propose that information theoretic principles can be used to evaluate interesting characteristics of time-ordered input sequences. In this paper, we present a novel data mining technique based on the Minimum Description Length principle that discovers interesting features in a time-ordered sequence. We discuss features of our real-time mining approach, show applications of the knowledge mined by the approach, and present a technique to bootstrap a decision maker from the mined patterns. Categories and Subject Descriptors Mining data streams, novel data mining algorithms, preprocessing and post processing for data mining, spatial and temporal data mining.