Font Size: 

In this paper, we propose a new clustering algorithm which is an improvement to a self-constructing clustering (SCC) method. The SCC processes all the data points incrementally. If the input data point is similar enough to an existing cluster, the point is added to the cluster. Otherwise, the data point forms a new cluster of its own. The method ends up with a set of clusters after it runs through the whole dataset once. However, once a data point is assigned to a cluster, there is no way to change the assignment afterwards. This may cause assignment errors and the efficacy of the clustering is reduced. In this paper, we adopt an iterative approach to overcome this shortcoming. A data point can be re-assigned to another cluster. Adding points into and removing points from a cluster are allowed to be done iteratively in the clustering process. The clustering work stops when all the assignments are stable, i.e., no assignment would be changed. The proposed approach can result in better clusters, and experimental results show that it performs better than SCC for real world datasets.

data mining, clustering, self-constructing clustering, similarity, classification.