Improved robust nonparallel support vector machines

Nonparallel Support Vector Machine (NPSVM) is a binary classification approach that combines the advantages of both support vector machine (SVM) and Twin SVM (TWSVM). It finds two nonparallel hyperplanes by solving two optimization problems such that each hyperplane is closer to one of the classes and as far as possible from the other class. To deal with data uncertainty, the chance-constrained Robust NPSVM (RNPSVM) is developed in López et al. (Neurocomputing 364:227–238, 2019) that improved model fit. In this paper, we propose an improved version of RNPSVM (IRNPSVM) by replacing the \(\epsilon -\)insensitive tube of each class by a chance constraint corresponding to its upper hyperplane while keeping its lower hyperplane. This results in reducing the number of missing data of the related class. It is reformulated as second-order cone programming problems. Experiments on both UCI and NDC datasets show that the improved model has better classification accuracy and its learning time is faster for the majority of the datasets compared to RNPSVM.