Universal consistency of twin support vector machines

Abstract A classification problem aims at constructing a best classifier with the smallest risk. When sample size approaches infinity, learning algorithms for are characterized by an asymptotical property, i.e., universal consistency. It plays crucial role in measuring construction of rules. consistent algorithm ensures that larger is, more accurately distribution samples could be reconstructed. Support vector machines (SVMs) regarded as one most important models binary problems. How to effectively extend SVMs twin support (TWSVMs) so improve performance has gained increasing interest many research areas recently. Many variants TWSVMs have been proposed and used practice. Thus this paper, we focus on consistency setting. We first give general framework TWSVM classifiers unifies Based it, then investigate TWSVMs. To do this, some useful definitions risk, Bayes risk Theoretical results indicate is valid various under certain conditions, including covering number, localized number stability. For applications our framework, several considered.