A NOVEL APPROACH FOR SEMI SUPERVISED CLUSTERING ALGORITHM

Semi-supervised clustering process is adopted for the improvement of the clustering performance, by considering the supervision of user in the form of pair wise constraints. This paper studies about the active learning problem of selecting pair wise cannot link and must-link constraints for semi-supervised clustering. The expansion of the neighborhoods is done by the active learning method, by selecting and querying the relationship of the informative points with the neighborhoods. Under this framework, the classic uncertainty based principle is built and a novel approach to evaluate the uncertainty related with each data point is presente. Furthermore, a selection criterion is introduced for the effective management of the amount of uncertainty of each data point with the expected number of queries required to resolve this uncertainty. So, that the selection of the queries that have the highest information rate is allowed. Evaluation of the proposed method on various benchmark data sets is performed. The experimental results demonstrate the consistent and substantial improvements of the proposed technique, with respect to the conventional state-of-the-art techniques