Research On Clustering Methods For High Dimensional Data And Their Applications

With the development of the Internet, data on the Web is increasing explosively. Clustering analysis is the basic method of Data Mining which has a wide range of applications. However, in the practical applications, the data is usually with high dimensions. Affected by the "Curse of dimensionality", traditional clustering approaches are often unable to obtain high-quality clustering when processing high-dimensional data.Dimension reduction and the robustness of clustering are two key issues in the field of clustering on high-dimensional data. How to reduce dimensions with good accuracy efficiently has become a hot research topic. Furthermore, the revelent methods for selecting the initial centers effectively and detect isolated points for high-dimensional data is another concerned problem as the robustness enhancement of the clustering algorithms.Furthering to the above two key problems, we focus on dimension reduction and initialization methods for high-dimensional clustering algorithms, as well as their applications in Information Security. The main contributions of this paper are as follows: (1) Based on Rough Set and modified Genetic Algorithm, a muti-filtering feature selection approach is proposed for intrusion detection. The significance of feature is used as heuristic information when initializing population, which can boost the convergence rate and obtain the optimized the results. (2) A local density based method is proposed to search for initial cluster centers on high-dimensional data. And this method is applied to Spam Detection and Intrusion Detection. Our approach can enhance the robustness including stability and reliability of high-dimensional clustering algorithms.