Research On Mining Outliers Of Stream Data Based On SPF

With the development of information technology, stream data emerge inmany applications. Example of stream data include network monitoring,meteorological monitoring, stock trading, train ticketing system, sensornetworks et al. Compared with traditional data, a data stream is an orderedsequence, which arrives in a multiple, continuous, rapid, unpredictable andtime-varying. Many algorithms become invild while mining data streams. It isvery important to research on algorithms of mining stream data.Outlier data exist widely in real applications, for example, a small amountof outlier data could indicate financial fraud in the bank transaction; real-timemonitoring outlier data could be used to observe the variation of diseases orprevent from the disease outbreaks; outliers detection could aviod the harm ofmachine fault in the airport security system; detecting effectively outlier datacan refrain from the diffusing of fallacious messages. At present, the research onoutlier detection has obtained some good achievements, which are based onstatistics, deviation, clustering, distance, density, and are difficult to be appliedin the occasions those require high detection accuracy and real-time, such asoutlier detection in stream data. This thesis studies the problem as follows. (1) Combined with the peculiarity factor and sliding window, a definiton ofoutlier based on window is proposed, through the analysis of mining on streamdata and outlier detection;(2) After analyzing the characteristics of data in the window, the algorithmof outlier detection based on optimal window is proposed, the learning methodof optimal window is designed, and they are applied to outlier detection instream data. Experiments on some data sets, the results show that the proposedmethod is effective and feasible;(3) To explore opinion detection according to the proposed method ofmining stream data.