Detecting Change in Data Stream: Using Sampling Technique

Detecting changes in data streams is an important area of research with many applications in which the underlying generating mechanism of data stream is constantly evolving, not liking the static data that was generated by a fixed process. To understand and study the evolving mechanism of the data stream, detecting the changes in data streams is necessary where the ´change´ is defined normally as ´the change of distribution of the streaming data´. In many real-world applications, this problem is challenging. Firstly, only the data stored in a fixed size of memory can be examined, whereas the changes might occur within data items with much bigger size. Moreover, the streaming data arrive at a high speed rate, so if we can not process the data quickly, the detecting of the change will be impossible. In this paper, we attack the issues by introducing a change detection method that utilizing sampling technique. A formal definition of the change in data stream will be given. The approach assumes that the points in the data stream are independently generated, but otherwise no assumptions on the nature of the generating process. The empirical study demonstrates the superiority of our techniques over the existing ones in terms of both efficiency and effectiveness.