Research On Spectral Cluster For Outlier Detection

With the constant development of database technology and Internet, the ability what people use information technology to obtain data and knowledge is substantially increase. Data mining is arising in this era ,it is essence a non-trivial process which obtain valid, novel, potentially useful and ultimately understandable patterns from databases, data warehouses or other repositories of large amounts of data. Outlier refer to data objects away from the conventional data, the data does not meet the general pattern and behavior. The majority of them showed significant difference with the conventional objects, so that suspicion may be caused by another mechanism. Therefore, as an important branch of data mining, outlier data mining has been applied to the medical, communications, electric power, finance, weather forecasting, intrusion detection and many other fields. At present, the outlier data mining has become a focus in the areas of machine learning, databases, experts and scholars. Outlier mining is divided into outlier detection and outlier analysis, the first stage is using some algorithm to detect outliers and the second stage is analysis outlier ,then access to knowledge. This article focuses on the first stage --outlier detect.Because spectral cluster can be clustered in any shape, gradually become a hot spot of today's clustering research. In this article, spectral clustering algorithm is successfully applied to the outlier data mining field, proposed outlier detect algorithm based on NJW, and Experimental results show the effectiveness of the algorithm. generally, this article focused primarily on the following areas:①Analysis of the relationship between data mining and outlier detect, then give the corresponding excavation process.②Introduced some outlier detect algorithm, Detailed comparison of their advantages, disadvantages and applicability.③Highlights the relevant theory of spectral clustering, and analysis its advantages④proposed outlier detect algorithm based on NJW, and Experimental results show the effectiveness of the algorithm.⑤Put forward the following research work in the main direction, and prospects the future development in outlier data mining.