Research On Improved Algorithm On Burst Detection In Data Streams

The burst detection in data streams is an important branch of research in the field of data steams and has been attracting more and more scholars' attention. The burst detection in data streams has wide variety of real world applications. For instance, it can be applied to financial, monitoring network traffic and other critical important areas. Existing burst detection methods mostly be focused on burst of monotonous accumulation function , which only be focused on fixed sliding windows. Moreover, the data structures of burst detection algorithms can not be adapted to real-time changeable characteristics of data streams. Hence, a real-time data structure with strong flexibility is needed , at the same time, a high-precision algorithm to detect burst in continuous and huge amount of data streams is necessary.In this thesis, based on the analysis of burst detection techniques, the burst detection algorithm and optimizing the input of data streams are improved. Our major contributions as follows:an improved algorithm of burst detection over data streams is proposed. Compared to the existing algorithm ,the algorithm presented in this thesis has the following advantages :The first, the algorithm of optimizing the data streams on the basis of Shifted Aggregation Tree is presented. So the algorithm can not be only relaxed the constraint conditions of the input in data streams, but also can be changed negative data streams or the positive and negative data streams into non-negative data streams, which can be handled the burst of non-monotonous aggregation function. The second, the data structure is constructed, it is a suitable elastic data structure that can be detected the burst in data streams efficiently. The last, the algorithm costs less time to detect burst in data streams, then improves the efficiency of burst detection, and reduces the time complexity.Theoretical analysis and experimental results show that the improved algorithm proposed in this thesis is suitable for burst detection over data streams. It can be concluded that the algorithm proposed in this paper can be achieved higher precision with less time complexity compared with existing research algorithms.