Solution Path Algorithm for Kernel Distributionally Robust Support Vector Machines

The distributionally robust optimization (DRO) method has emerged as a robust and effective framework for addressing the challenges posed by data distributional uncertainty in machine learning problems. This paper presents a kernel distributionally robust support vector machine model, which integrates the kernel and DRO methods with the capability of capturing complex, nonlinear decision boundaries, and making robust classifications to a wide array of real-world applications. A solution path algorithm is proposed to efficiently solve this model and speed up the hyperparameter tuning. Numerical experiments are performed on different datasets to validate the efficiency and effectiveness of the proposed approaches.