Reseach On Semi-Supervised Clustering Algorithm Based Convex Hull

Semi-Supervised Clustering Algorithm is a research focus in the international field of machine learning and data mining, it attracted many scholars to study in this field, and have achieved some results..In this paper, the Semi-Supervised Clustering Algorithm in data mining are studied, and a new semi-supervised clustering algorithm based convex hull is proposed. The clustering results are verified by experiments.This paper, to a certain extent, riches semi-supervised clustering algorithm for the research contens. It provises a new train of thought for the semi-supervised clustering problem solving, and has a certain scientific value and potential application.From the standpoint of machine learning, the traditional clustering belongs to the unsupervised learning, but semi-supervised clustering belongs to semi-supervised learing. In the process of semi-supervised clustering, the sample datas with lables or constrained informations are auxiliary for the processing of clustering. So, the most important problems in semi-supervise clustering is how to use the sample datas with labels or constrained information effectively to guide the clustering process for getting the more better result clusters. At the present time, semi-supervised clustering algorithms that have been well know can be roughly divided into two categories. One is the method based on limited, and the other is the method based on distance. The SCBCH proposed in this paper belongs the latter method.From the structure, this paper introduces the related theory about clustering firstly, the basic concepts in clustering are explaind. Due to the SCBCH algorithm belongs to clustering, therefore, these concepts are also suitable for SCBCH algorithm. Meanwhile, common types of clustering cluster and common clustering algorithm are also introduced. In subsequent content, this paper focuses on the revelant knowledges about semi-supervised clustering, such as the background and meaning of studying semi-supervised clustering, the basic theory and development situation of semi-supervised clustering, and common algorithms of semi-supervised clustering. They are all used to establish the theoretical foundation for the SCBCH algorithm proposed later in this paper. In second chapter, the paper introduces some kinds of measure methods for clustering clusters, include their principles. These established good basis for the valid results getting by verification experiments on SCBCH algorithm.Secondly, the third chapter in this paper introduces the related theories and common algorithm about convex hull, and the discusses reasons that introduce the convex hull into the semi-supervised clustering. Then the paper shows the explore experiments about convex hull algorithm's time complexity in different dimension. Those experiments show that, convex hull algorithm is suitable in low dimension space. All above provide the support on the theory for the verification experiments of SCBCH algorithm.In the fourth chapter, this paper proposes the core content, SCBCH algorithm. In this chapter, the paper shows the main component of the algorithm from three directions in details, include the meanings, process, varification experiments of it. Those experiments show that, compared with traditional clustering, SCBCH algorithm has a certain promotion both on clustering accuracy and compactedness between clustering clusters, according with the characteristics of semi-supervised clustering. Meanwhile, SCBCH algorithm has much fewer on the effects compared to Seeded-KMeans, a kind of well-known semi-supervised clustering algorithm proposed by Sugato Basu. So the validation of the algorithm is proved.At last summarized to the full paper, and indicated the next work. Then proposed the own views about the further development of the semi-supervised clustering based convex hull algorithm.