Research On Intrgrated Management Of Outlier In Intelligent Data Analysis

Data is one of critical resources of information era. It's quality has important influence on the result of intelligent data analysis, especially the emergence of outlier. A strange data value that stands out because it is not like the rest of the data is called an outlier. It doesn't accord with common data patterns and has different engendering mechanism.There are two types of outlier. The first is real outlier which includes significant knowledge from application aspects. The second is noise outlier which is measurement or recording errors and it may lead to wrong analysis result. There are two types of attitude toward outlier. One is to reject all outliers. Although rejection may reduce misleading effct, the significant information may be dropped. The second attitude is including. Though including may keep any information, the analysis result may be misleaded. Researcher's attention is mainly focusing on the detection of outlier. There are no comprehensive management idea about it. This dissertation proposes an integrated management method about outlier in intelligent data analysis. It describes handling way from outlier detection,discrimination and discovering knowledge from real outlier. To exert the integrated method effectively, it also introduces a process to reduce noise outlier. The innovation point of this dissertation is focusing on the construction of integrated management method about outlier. It can be divided into three parts: Introduce the method of detecting outlier by Self Organizing Map on the base of comparing existing detection methods of outlier;Propose two complementary strategies to distinguish different types of outlier by using domain knowledge;Put forward an interactive-heuristic knowledge discovery process; To help understand the tenor of this dissertation, it also splits an application and explains it in correlative chapters.