Mining data streams with periodically changing distributions

10 years 4 months ago
Mining data streams with periodically changing distributions
Dynamic data streams are those whose underlying distribution changes over time. They occur in a number of application domains, and mining them is important for these applications. Coupled with the unboundedness and high arrival rates of data streams, the dynamism of the underlying distribution makes data mining challenging. In this paper, we focus on a large class of dynamic streams that exhibit periodicity in distribution changes. We propose a framework, called DMM, for mining this class of streams that includes a new change detection technique and a novel match-andreuse approach. Once a distribution change is detected, we compare the new distribution with a set of historically observed distribution patterns and use the mining results from the past if a match is detected. Since, for two highly similar distributions, their mining results should also present high similarity, by matching and reusing existing mining results, the overall stream mining efficiency is improved while the accu...
Yingying Tao, M. Tamer Özsu
Added 02 Sep 2010
Updated 02 Sep 2010
Type Conference
Year 2009
Where CIKM
Authors Yingying Tao, M. Tamer Özsu
Comments (0)