Modified Newton’s method for solving parametric ν-support vector regression with Universum data

Universum, representing a third category distinct from the two primary classes in classification tasks, facilitates the incorporation of prior knowledge into the learning process. Extensive studies have confirmed its effectiveness in improving both supervised and semi-supervised classification models. Recently, Universum data has been introduced into parametric \nu-support vector regression (UPar-\nu-SVR) to enhance generalization performance. In this paper, we present a Newton-based method for solving UPar-\nu-SVR, with the objective of further improving its efficiency and accuracy. Our approach reformulates the problem into an unconstrained convex optimization framework and employs a generalized Newton’s method for its solution. To assess the effectiveness of our proposed method, we conduct comprehensive experiments on multiple UCI benchmark data sets. The experimental results indicate that our algorithm outperforms existing techniques, providing superior generalization capabilities and computational efficiency.

کلیدواژه ها:

Parametric \nu-support vector regression ، Univesum data ، Newton's method