Sparse Pinball Twin Parametric Margin Support Vector Machine

Urairat Deepan, Poom Kumam, Parin Chaipunya

Abstract

The main purpose of this paper is to construct a twin parametric margin support vector machine combined an $\epsilon-$insensitive loss function for finding a pair of parametric margin hyperplanes that automatically adapts to the parametric noise with arbitrary shape to capture the data structure more accurately. We exhaustively test several UCI datasets demonstrates that our SPTPMSVM is noise insensitive, retains sparsity in most cases. Finally, we present the numerical experiment and compare our model with other models.