ABSTRACT Although there are some recent methods, such as SVR, TSVR, and TPSVR, that enhance training speed by using nonparallel hyperplanes, they remain sensitive to outliers and non‐Gaussian noise, like the ‐insensitive loss function‐based methods. We propose a method to tackle this issue, namely the Huberized TPSVR framework. This model offers a dual protection mechanism: the parametric margins adapt to the local density in the data, while the integrated Huber loss function guides the learning process through robust noise and extreme outliers. Our experiments on complex chemical and clinical data reveal the remarkable superiority of Huberized TPSVR over standard models in terms of both stability and accuracy, which provides an efficient approach for addressing the messy nature of real‐world multidimensional data.
Deepan et al. (Thu,) studied this question.