Outlier Detection Technic On Probilistic Stream

Outlier detection is an important technology in data mining, with value in many application fields such as network intrusion detection, event detection in wireless sensor network (WSN) and so on. Outlier detection has been learned deeply in deterministic data; however, it is a new research field in probabilitic data. Probabilistic stream is an emerging data model in recent years, with the both characters of probabilitic data and data stream. Probabilitic data describes the uncertainty of objects, better reflecting the real-world, enhancing feasibility of some applications; however, it also brings some challenges to data management. Thus, probabilitic stream management not only needs to meet the requirement of data stream, but also solve the problems brought by uncertainty.This paper firstly brings forward outlier detetion on probilistic stream and takes the preliminary exploration in distance-based outlier dectection on probabilistic stream, in which probabilistic stream is destribed by sliding window model which is widely used in data stream management and independent uncertain tuples use discrete probability value to denote their existence. For the uncertainty, the original definition for distance-based outlier on deterministic data is disabled, so this paper begins with the new definition for distance-based outlier on probabilitic data, with probability as its measurement.Detection on single window is the base for the detection on the whole stream, getting the detection result for snapshot of each window. Refering previous work, a pruning principle is found as the base of PDA method, which solves the problem of detection on single window. Then, WPDA is brought forward, using PDA on sliding window, as the preliminary method for outlier detection on probilistic stream. It is an important character of sliding window to increamtally maintaining the information in the window. This paper brings forward mechanism of increment maintenance, avoiding repeated processing due to discarding pruning imformation as WPDA, and brings forward OWPDA method based on this mechanism.The experiment tests OWDPA and WPDA and analyzes the testing results. As the testing results show, OWPDA method well enhances the effiency through mechanism of increment maintenance.