Resolution of the Probabilistic Vector Machine Problem via Single Linear Program

Abstract

This paper presents a significantly improved version to a recently introduced hyperplane classifier, Probabilistic Vector Machine (PVM). The main goal is to provide a formulation which allows fast and robust resolution of the classification problem as approached by the PVM algorithm. The main result is the introduction of a single linear program (LP) form which avoids the iterative process initially introduced by PVM. This allows comparison to state of the art algorithms such as Least Squares Twin Support Vector Machines(LSTSVM) and Robust Twin Support Vector Machines (R-TSVM). The results prove that PVM is both highly competitive and stable.