Percentage-Error Support Vector Regression R package psvr version 0.1.0 Implements four support vector regression (SVR) models derived from a unified mathematical framework for percentage-error loss functions: epsilon-SVR minimizing the mean absolute percentage error (MAPE), its symmetric kernel extension, least-squares SVR (LS-SVR) minimizing the root mean square percentage error (RMSPE), and its symmetric counterpart. All models require strictly positive targets. The epsilon-SVR models are solved via a built-in sequential minimal optimization (SMO) algorithm (with 'osqp' available as an optional alternative backend) and the LS-SVR models via a linear system (base R). See Benavides-Herrera et al. (2026) <doi:10.3390/math14101679> for the mathematical derivations. Software Depends: R (>= 4.1) Imports: dials, Matrix, parsnip, Rcpp (>= 1.0.10), rlang, scales, stats, tune, utils, workflowsets LinkingTo: Rcpp Suggests: dplyr, ggplot2, knitr, osqp, recipes, rmarkdown, rsample, testthat (>= 3.0.0), tibble, workflows, yardstick Pablo Benavides-Herrera <pbenavides@iteso.mx> Comprehensive R Archive Network (CRAN) Pablo Benavides-Herrera [aut, cre] (ORCID: <https://orcid.org/0000-0003-4926-4763>) MIT + file LICENSE (https://CRAN.R-project.org/package=psvr/LICENSE) 2026-09-24 application/tgz https://CRAN.R-project.org/package=psvr doi:10.32614/CRAN.package.psvr