An Enhanced Projection Twin SVM Model for Classification

By taking the L0/1-soft-margin loss and the working set selection strategy into account, we establish an enhanced projection twin SVM optimization model for general classification problems. The optimality properties of the presented model are analyzed via proximal stationary point theory. A working-set ADMM-type algorithm with quadratic correction terms is further developed for efficient model solution. Numerical experiments on synthetic samples, UCI benchmarks, and NDC datasets with different sample sizes illustrate the promising performance of the proposed method in comparison with existing alternatives.