Mining Pattern Changes in Sensor Data Streams usingA pproximate Sequence Alignment

摘要

In a typical surveillance scenario, a system of multiple sensors is used to detect emitters in a particular geographical area oj interest. The collected emitter data is then further processed and analyzed to determine higher level information and intelligence. We have developed a data mining system which is able to process streams of emitter data to determine which emitters are of interest, and the significant events and observations of those emitters. In particular, we have focused on detecting slight changes in the occurrence behavior of emitters which may be indicators of more significant events. We have chosen to leverage approximate sequence alignment techniques to determine these changes by viewing emitter behaviors as a sequence of characters indicating their occurrence over time.