l1-norm Nonparallel Support Vector Machine for PU Learning

PU (Positive and Unlabeled) learning problems are defined on a dataset that contains a small number of positive examples and a large number of unlabeled examples. Recently nonparallel support vector machine (NPSVM) has been applied to solve PU learning problems. In this paper we propose to use the l₁ norm regularization term in NPSVM for solving PU learning problems. Two linear programming problems are constructed to get the solution. Preliminary experimental results are reported to show the usefulness of the proposed l₁-NPSVM in both classification and feature selection.