Kernel-based parametric validity index for assessing clusters from microarray gene expression data

In this paper, we develop a kernel-based parametric validity index (KPVI), which not only inherits robust feature from the newly proposed PVI, but possesses extra superiority inherited from the kernel method. The KPVI employs the kernel method to calculate both the inter-cluster and the intra cluster dissimilarities. Furthermore, we develop several rules to guide the selection of parameter values by examining the dissimilarity densities of different datasets such that the maximal appropriate values of the parameters for individual dataset can be obtained. We evaluate the new KPVI for assessing five clustering algorithms in both synthetic and real gene expression datasets. The experimental results support that the KPVI has the most superior performance among the existing validation algorithms, even better than the PVI.