Abstract
Multi-clustering is a technique for amalgamating the results of many runs of a standard clustering algorithm to obtain a clustering of data which avoid artifacts introduced by the underlying metric. Multi-clustering also yields an advisory, called a cut plot, as to the number of "natural" clusters present in the data. In order to perform multi-clustering a number of parameters must be chosen. This paper tests evolutionary algorithms that perform parameter setting for multi-clustering on synthetic data set with designed numbers of clusters. A evolutionary algorithm and an evolution strategy are compared. The superior algorithm, the ES, is then used to set parameters for four microarray-like data sets. Evolutionary parameter setting is found to more than double the range in which the cut plot detects the correct number of clusters when compared to hand-chosen parameters arrived at by serial parameter optimization. This paper also presents a new technique for accelerating multi-clustering, iteration limiting, and demonstrates that the technique may be implemented to speed up multi-clustering without impairing performance. The evolutionary results support the use of iteration limiting in multi-clustering
Index
Terms
Available to subscribers and IEEE members.
References
Available to subscribers and IEEE members.
Citing Documents
Available to subscribers and IEEE members.
Cart
