A comparative review of graph-based ensemble clustering as transformation methods for microarray data classification

Recently, the use of ensemble data matrix as a transformed space for classification has been put forward. Specific to the problem of predicting student dropout, the matrix generated as part of summarizing members in a cluster ensemble is investigated with a number of conventional classification methods. Despite the reported success in comparison to the case of original data and other attribute reduction techniques like PCA and KPCA, the study is limited to only one ensemble matrix that is created by the link-based ensemble approach or LCE. To provide an comparative review with respect to the aforementioned problem, this paper includes the experiments and findings obtained from the use of different graph-based ensemble algorithms as data transformation methods for microarray data classification. The empirical study can be hugely useful particularly for those working in bioinformatics, and generally applicable to any classification problem. Besides, the review initiates another interesting challenge for many researchers in the field of cluster ensemble to coupling their models with this hybrid, clustering-classification learning.