Advancing data clustering via projective clustering ensembles

摘要: Projective Clustering Ensembles (PCE) are a very recent advance in data clustering research which combines the two powerful tools of ensembles and projective clustering.Specifically, PCE enables ensemble methods to handle composed by solutions. has been formalized as an optimization problem with either two-objective or single-objective function. Two-objective shown generally produce more accurate results than its counterpart, although it can object-based feature-based cluster representations only independently one other. Moreover, both early formulations do not follow any standard approaches ensembles, namely instance-based, cluster-based, hybrid. In this paper, we propose alternative formulation overcomes above issues. We investigate drawbacks define new problem. This is capable treating object- whole, essentially tying them distance computation between solution given ensemble. cluster-based algorithms for computing approximations proposed formulation, have common merit conforming ensembles. Experiments on benchmark datasets significance our heuristics outperform existing methods.