The Study Of Clustering Algorithm Based On Density

Density-based clustering method plays an important role in cluster analysis.It has been widely used in financial,marketing,information retrieval,information filtering, scientific observation and engineering.It is the focus of current research.Density-bas-ed clustering algorithms are studied in this paper,and a local scaling based clustering algorithm is proposed.Clustering algorithm based on grid and density integrates density-based and grid-based clustering algorithm,and combines advantages of two clustering algorithms.This paper presents a improved clustering algorithm based on grid and density.The main contents are outlined as following:(1) At first,an introduction of data mining techniques and the overview of cluster analysis techniques are concerned.Then,the paper descirbes basic clustering concept and data structure,and specifies some commonly used clustering algorithms and data preprocessing methods.(2) Parameter sensitivity and unsatisfactory performance in clustering uneven density distribution datasets are the main weekness of DBSCAN.the paper presents a improved local scaling based clustering algorithm to improve the weekness of DBSCAN.The algorithm uses the regional distribution to measure density of the point, candidate core point is defined to improve the search efficiency of clustering.Outlier detection method LOF is used to detect outliers.(3) The time performance of algorithm based on grid and density are independent of the size of dataset. A improved clustering algorithm based on grid and density is proposed in the paper.Dataset is mapped to a grid structure by density function,then grid segmentation technique is used to binarize the grid structure.The algorithm cluster the density connected region on the binarized grid structure.(4) An intrusion dectection model is proposed based on density-based cluster analysis.The local scaling based clustering algorithm is used to train the intrusion knowledge base. Experimental results show the effectiveness of local scaling based clustering algorithm in application.Experimental results show that the clustering result by local scaling based clustering algorithm,with a new density function and local scaling,is better than that of DBSCAN for clustering uneven density distribution datasets.The sensitivity of parameters have also been enhanced.Clustering algorithm based on grid and density is the reinforcement of density based clustering algorithm,for its ability to clustering uneven density distribution datasets and independence of time performance with the size of datasets.