Detection of Concept Drift for Learning from Stream Data

In data processing under dynamic environment such as stream, the time is one of the most significant facts not only because the size of data is dramatically increased but also because the context of data could be varied over time. To learn effectively from dynamic data evolving over time, it is required to detect the drift of the concept of data. We present a method to detect it by utilizing the correlation information of value distribution and apply our method to a learning task on a multi-stream data model. The result of experiments on a synthetic data set shows that our approach could provide a reasonable threshold to detect the change between windowed batches of stream data.