Energy-based structural least squares twin support vector clustering

Abstract

Clustering is an unsupervised learning algorithm and it is widely used in machine learning. Twin support vector clustering (TWSVC) is a new plane-based clustering algorithm, which exploits information from both within and between clusters to generate plane for each cluster. However, TWSVC suffered from relatively poor performance because it ignores the intrinsic structural information of data and solves a series of quadratic programming problems (QPPs). In order to address these problems, in this paper, we propose a novel energy-based structural least squares twin support vector clustering, termed as ESLSTWSVC. Firstly, based on least squares twin support vector clustering (LSTWSVC), we introduce within-class covariance matrix into the objective function of LSTWSVC to obtain more intrinsic structural information. ESLSTWSVC solves a series of system of linear equations rather than to solve QPPs in TWSVC, it leads to simple algorithm and less computation time. In addition, ESLSTWSVC converts the constraints of LSTWSVC into energy-based model by introducing an energy parameter for each cluster that makes ESLSTWSVC more robust. Experiments are performed on artificial datasets as well as UCI datasets, and the experimental results illustrate the effectiveness of the proposed algorithm.