Improvements And Applications Of Cluster Analysis Algorithm CLIQUE

With the rapid development of computer technology, the rapidly growing popularity of the network, it is production, the ability to collect data continuously improve, the amount of data at an unprecedented rate of growth in mass. The face of huge amounts of data, how to extract data from the mass of information, knowledge resources, so as to avoid "data rich and poor knowledge of important" situation, has become a pressing need to be addressed.Data Mining is to solve the problem of technology. The field of data mining as a major area of technology, clustering analysis is similar to the target is divided into clusters, which helps people search and find useful information and knowledge.CLIQUE algorithm is based on the mesh density and the clustering method. In a large high-dimensional data sets, the clustering algorithm is better, but because of its sub-space pruning method is simple, the use of hardened unit grid defect classification, leading to its efficiency and clustering quality is not high enough.In response to these problems, this thises, an improved algorithm CLIQUE. The basic idea of the new algorithm is bound by the conditions of the monotony of the same nature CLIQUE algorithms combine together for the clustering of the candidate of "pruning" operation to reduce the CLIQUE algorithm search of the "blindness"; the use of adaptive mesh technology-intensive units significantly reduce the candidate set of input, at the same time, it reduces the need to deal with the size of data sets; boundary adjustment of technology to enhance the use of clustering accuracy.In order to prove the advanced nature of the new algorithm, UCI data sets used in this thesis two algorithms on the experimental results show that the new clustering algorithm is faster, more scalable, better clustering quality.Finally, the new algorithm in intrusion detection systems. Intrusion Detection in KDDCUP99 data sets, Application of the new algorithm, respectively, of various types of attacks on the experimental data sets, and then attack on the mixed types of experimental data sets. The results show that the new algorithm other than the current algorithm has better clustering quality.