Semi-supervised Uncorrelated Feature Selection

EasyChair Preprint 1551

Abstract

In this paper, we propose an uncorrelated feature selection method for semi-supervised feature selection task, namely SSUFS. The new method extends the Rescaled Linear Square Regression by imposing an Uncorrelated Regularization (UR) to select only a small number of important features from highly correlated features. With this regularization, the new method is able to select lowly-nonlinear-correlated important features. SSUFS was compared with 5 feature selection methods on 5 datasets and the experimental results show the superior performance of the new method.

Keyphrases: Nonlinear Uncorrelated Feature Selection, feature selection, semi-supervised learning

@booklet{EasyChair:1551,
  author    = {Weichan Zhong and Xiaojun Chen and Feiping Nie},
  title     = {Semi-supervised Uncorrelated Feature Selection},
  howpublished = {EasyChair Preprint 1551},
  year      = {EasyChair, 2019}}