Abstract
For classification problems, twin support vector machine with pinball loss (Pin-GTSVM) is noise insensitive and has better performance than twin support vector machine (TWSVM). However, it lacks sparsity in comparison to TWSVM. In this article, to maintain a trade-off between the noise insensitivity and sparsity of the model along with preserving the theoretical properties of pinball loss, we propose universum twin support vector machine with truncated pinball loss (Tpin-UTWSVM). The proposed Tpin-UTWSVM considers universum data which gives prior information about the distribution of the data, thus improves the generalization performance of the proposed model. Further, the proposed optimization problem is non-convex and non-differentiable which is solved by concave–convex procedure. We employed the SOR approach to train the proposed model effectively with minimum training time. We conducted numerical experiments on 19 UCI binary datasets with different noise levels to validate the noise insensitivity of the proposed Tpin-UTWSVM model. We also conducted numerical experiments for electroencephalogram (EEG) signal classification and Alzheimer’s disease (AD) detection. The overall experimental outcomes and statistical tests demonstrate the superiority of the proposed Tpin-UTWSVM model in comparison to the baseline models. The source code for the proposed Tpin-UTWSVM is available at https://github.com/mtanveer1/Universum-twin-SVM-with-truncated-pinball-loss.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have