Twin pinball loss support vector hyper-sphere classifier for pattern recognition

Motivated by twin support vector hyper-sphere (TSVH) and support vector machine with pinball loss (pin-SVM), this papaer formulated a twin pinball loss support vector hyper-sphere (TPSVH) classifier. TPSVH is obtained by introducing pinball loss with quantile distance into TSVH. It has TSVH-type objective function and similar inequality constraints with pin-SVM. And it is stable and insensitive to noise samples, which makes TPSVH fitter for pattern classification problem. From the classification experiments for synthetic and UCI datasets, it can be clearly seen that TPSVH has better classificaiton accurary and generalization performance compared with other classifiers. In a word, the novel classifier proposed in this paper enjoys excellent performance in stability and insensitivity to noise.