Study On The Three-way Cluster Ensemble Approach Based On Icremental Ensemble Members

In the terminology of machine learning, cluster analysis is an instance of unsupervised learning. Clustering involves grouping data into classes based on some measure of inherent similarity or distance. After grouping, objects in the same group(called a cluster) are more similar(in some sense or another) to each other than to those in other groups(clusters). It has been applied in many fields such as artificial intelligence, image processing, research on market and so on. This thesis mainly focuses on the problem of cluster ensembles.The main contrbutions of the thesis is as follows.First, a representation of clustering is proposed based on the interval sets. In the thesis, a cluster is presented by an interval set which has the upper and lower bound inspired by the three-way decisions. Then, a cluster is divided naturally into three regions, the positive region, the boundary region and negative region. Objects in the positive region belong to the cluster definitely, objects in the negative region do not belong to the cluster definitely and objects in the boundary region might belong to the cluster or not. The expression can express the two-way results as well as the three-way results.Second, a voting cluster ensemble approach is presented based on the three-way decisions. Cluster ensemble can combine the outcomes of several clusterings to a single clustering that agrees as much as possible with the input clusterings and improves the stability of the clustering algorithm. Since the three-way decision theory is widely studied recently with its advantages of human cognition and decision-making, the main objective of this paper is to propose a transfer approach based on the framework of cluster ensemble. The proposed approach gets different hard partitions through ensemble members which are the existing clustering algorithms. Then, it decides some objects definitely to the corresponding clusters by matching the clusters' tag and intersection operations. Finally, it decides the rest objects by three-way decisions based on voting. The preliminary experimental results show that the proposed approach is effective.Third, a novel efficient gradual three-way decision cluster ensemble model is proposed, which adopts the idea of cluster core. In this thesis, a novel efficient gradual cluster ensemble model using the three-way decision theory is devised, which has the ability to deal with both two-way decision clustering and three-way decision clustering. The concept of cluster core is introduced to improve the efficiency of the algorithm, which reflects the minimal granularity distribution structure agreed by all the ensemble members. Besides, the idea of gradually decision making is adopted. That is, the boundary region objects are clustered again based on the incremental ensemble members until all the objects are clearly identified. So that, we can get two-way or three-way decision final clusters. The experimental results using the artificial 2D data sets and the multi-dimensional real-world data sets with different shapes validate the effectiveness of the proposed approach. The comparative experimental results also show that the proposed approach has a lower time cost and does not sacrifice the accuracy.