Research On Twin Support Vector Regression by Hiker Optimization Algorithm

Abstract

This paper presents an enhanced twin support vector regression (TSVR) model integrated with the hiker optimization algorithm (HOA) and successive 0ver-relaxation (SOR) method to address challenges in regression accuracy and computational efficiency. HOA is employed to optimize hyperparameters, while SOR accelerates the solution of quadratic programming problems by reducing memory usage for sparse matrices. Experimental results on 8 synthetic test functions with diverse noise types and 18 benchmark datasets demonstrate that the proposed HOA-TSVR significantly outperforms traditional models. However, performance fluctuations on high-dimensional small-sample datasets highlight sensitivities to data distribution and feature redundancy. The study concludes that the hybrid approach enhances generalization and robustness, offering a promising framework for complex data modeling, with future research directions focusing on adaptive feature selection and multi-scenario applications.