PROJECT TITLE :

Adaptive ensembling of semi-supervised clustering solutions - 2017

ABSTRACT:

Conventional semi-supervised clustering approaches have many shortcomings, such as (one) not absolutely utilizing all useful must-link and cannot-link constraints, (two) not considering how to accommodate high dimensional knowledge with noise, and (3) not fully addressing the requirement to use an adaptive method to more improve the performance of the algorithm. In this paper, we have a tendency to first propose the transitive closure primarily based constraint propagation approach, which makes use of the transitive closure operator and also the affinity propagation to handle the primary limitation. Then, the random subspace based mostly semi-supervised clustering ensemble framework with a collection of proposed confidence factors is intended to handle the second limitation and give additional stable, robust, and accurate results. Next, the adaptive semi-supervised clustering ensemble framework is proposed to deal with the third limitation, which adopts a newly designed adaptive method to look for the optimal subspace set. Finally, we adopt a collection of nonparametric tests to compare totally different semi-supervised clustering ensemble approaches over multiple datasets. The experimental results on twenty real high dimensional cancer datasets with noisy genes and 10 datasets from UCI datasets and KEEL datasets show that (one) The proposed approaches work well on most of the $64000-world datasets. (a pair of) It outperforms other state-of-the-art approaches on 12 out of 20 cancer datasets, and eight out of 10 UCI machine learning datasets.