Study On Path-Based Clustering Algorithm Of Partition

With the rapid development of information science and technology, the amount of information around the world dramatic increasing. It is a challenge task that we get the valuable information from these complex data. Cluster analysis is an important component in data mining, and is an important tool in data partition or classifying process. Cluster analysis has been widely applied in market analysis, image processing, web document classification, etc.We analysis the state of the art in clustering algorithms and focus on study path-based clustering algorithms of partition in this thesis, which based on the detailed knowledge of cluster analysis. The main contents of this thesis are as follows.(1) Traditional K-means clustering algorithm is sensitive to the select of initial clustering centers and isolated point. Considering these problems, a new method based on density of point is presented in chapter 3. This method can select of the initial cluster centers more efficient in the process of clustering. Isolated points are treatmented specially which doesn't affect the value of cluster center with the iterative method. Theory and experimental results demonstrate that K-means algorithm based on optimal initial clustering center is better than the traditional K-means algorithm.(2) Traditional partition clustering algorithm can't discover clusters of arbitrary shapes. For this problem, a new path-based similarity measure called distance metric is proposed, and a new objective criterion function is designed in chapter 4. A path-based clustering algorithm of partition is proposed consequently. Experimental results show that our method can determine the number of clusters automatically, and obtain the ideal cluster results. It is insensitive to isolated point.