Pinball twin bounded support vector clustering

Abstract

Unsupervised machine learning algorithms are extensively used for unlabelled data clustering. Twin support vector clustering (TWSVC) and twin bounded support vector clustering (TBSVC), plane-based clustering algorithms introduced recently, work on twin support vector machine (TWSVM) principles and are used in widespread clustering problems. However, both TWSVC and TBSVC are sensitive to noise and suffers from low re-sampling stability due to usage of hinge loss. The pinball loss, an alternative loss function first introduced in the pinball twin support vector clustering (pinTSVC) which makes the algorithm noise insensitive and more stable for re-sampling of data. In this paper, we propose an efficient plane-based clustering method called twin bounded support vector clustering using pinball-loss (pinTBSVC). The proposed pinTBSVC inherits various attributes from previous plane-based clustering algorithms (like the regularisation term or the structural risk minimisation principle) and improves upon them by implementing pinball loss for its optimisation. Unlike TWSVC and pinTSVC, the matrix appear in the dual formulation of the proposed pinTBSVC is nonsingular. Experimental results performed on several benchmark UCI datasets indicate that the proposed pinTBSVC outperforms TWSVC, TBSVC and pinTSVC. Furthermore, we also discuss the application of the proposed pinTBSVC in biomedical datasets. Numerical experiments show that the proposed pinTBSVC has shown better generalization performance on both synthetic and real-world benchmark datasets. The MATLAB implementation of the proposed pinTBSVC is available on https://github.com/mtanveer1. © 2021 IEEE