TWSVC+: Improved twin support vector machine-based clustering

Abstract

Based on twin support vector machines (TWSVM) model, the twin support vector clustering (TWSVC) is a planar clustering model that increases inter-cluster separation. Because the TWSVC is not a standard model for some variables, its solving algorithm consumes lots of time and does not always converge to the optimal solution. To solve the problem, this paper proposes a novel clustering model, denoted as TWSVC+, based on twin support vector machines (TWSVM). The TWSVC+ is convex and standard with respect to each variable. Therefore, it is possible to solve this model rapidly with an algorithm that converges to a global optimal solution relative to each variable. The author presented linear TWSVC+ and non-linear TWSVC+ for clustering linear separable clusters and linear inseparable clusters, respectively. Experimental results on real datasets of UCI repository show that the TWSVC+ was better than TWSVC and support vector clustering (SVC) in accuracy and training time.