A clustering ensemble framework based on system energy theory

A clustering ensemble method has been shown to be very effective inimproving both the performance and constancy of unsupervised classificationsolutions. However, it is difficult to find a consensus clustering result frommultiple clustering partitions. In this paper, we present a, new clusteringstrategy with system, energy theory. The system, composing of clusteringensemble members and, ensemble result, is the most stable when the clusteringensemble result is optimal, which system, energy is lowest. On the basis, a, newclustering ensemble framework and, the optimal object function are proposed, byusing a, system, energy function as as consensus function. Then, a, heuristicalgorithm, K-ETMC is designed, which is easy to implement, requiring multiplehash tables as the major data structure. The experiment shows that clusteringeffect is better, which can effectively improve clustering result. ICIC International © 2010.