Density Clustering Based SVM and Its Application to Polyadenylation Signals

Support vector machines (SVM) have been promising methods for classification analysis due to their solid mathematical foundations. Clustering-based SVMs are used to solve large samples classification problems and reduce the computational cost. In this paper, we present a density clustering based SVM(DCB-SVM) method to predict polyadenylation signal (PAS) in human DNA and mRNA sequences. We decrease the original data scale by using the density restricted hierarchical clustering. This strategy leads to solving smaller sized problems, making DCB-SVM work faster than standard SVM. According to the results of the PAS experiment, the proposed method is not only fast, but also shows better improvement in sensitivity than the SVM.