Newton-based approach to solving K-SVCR and Twin-KSVC multi-class classification in the primal space

Hossein Moosaei, Milan Hladík, Mohamad Razzaghi, Saeed Ketabchi

Abstract: Highlights•Two approaches, NK-SVCR and NTW-KSVC, for multi-class classification are proposed.•The proposed approaches solve unconstrained minimization problems in primal space.•NK-SVCR and NTW-KSVC use “1-vs-1-vs-rest” on training data for ternary outputs {−1,0,+1}.•The fast Newton-Armijo technique is modified due to once-differentiable unconstrained minimization.•We compare complexities, showing our methods need less effort than existing approaches.•Experiments assess strategies on artificial, USPS, and UCI data, showcasing their efficacy.