Research On Clustering Analysis Based On Cultural Algorithms

As a basic means of information processing, clustering analysis technology has become people's concern in recent years. Clustering analysis has also gained a wide range of research and application in machine learning, pattern recognition, data mining, information retrieval and many other fields.Clustering analysis is an important part of the Data Mining research. Clustering is the process of grouping the physical or the abstract object set into classes or clusters, so that the objects within the same cluster have high similarity in comparison to one another, but low similarity in different clusters.Cultural Algorithms is a new evolutionary model. The algorithm is dual inheritance systems that besides the population component which traditional evolutionary computation methods have, there is an additional peer component belief space and a supporting communication mechanism between those two components.This paper proposes a new hybrid clustering algorithm, this algorithm takes the Cultural Algorithms as a frame, uses the K-Means Algorithm as clustering model, and designs belief space, population space, accept function and influence function for special clustering problem. First, Genetic Algorithm was used in population space, at the same time, Situational Knowledge, Normative Knowledge, and Topographic Knowledge were used in belief space for guidance, so the algorithm named KCAGA was presented. Then, Evolutionary Programming was used in population space so that the algorithm named KCAEP was put forward. Based on different belief space, influence function, KCAEP was divided into different versions: KCAEPⅠ, KCAEPⅡ, KCAEPⅠ', KCAEPⅡ', KCAEPⅢ, and KCAEPⅣ.Take research on belief space, experiments show that three kinds of knowledge such as Situational Knowledge, Normative Knowledge, and Topographic Knowledge used in belief space for guidance can result in more efficient clustering results, which are better than both Situational Knowledge and Normative Knowledge used. And the improved influence function can avoid the clustering results trapping in local minima.Experiments show that this algorithm can not only avoid the disadvantages of the classical K-Means clustering algorithm, but also have greater searching capability globally. The new algorithm achieved good results apply to resolve the clustering problem.