Collinear groupwise feature selection via discrete fusion group regression

作者：

Highlights：

• A collinear group-wise feature selection method was proposed

• The method adopts a fusion group constraint to reduce the variance of coefficients.

• The modified discrete first-order method was used to obtain near optimal solutions.

• Comparison studies show that the proposed method outperformed existing methods.

摘要

•A collinear group-wise feature selection method was proposed•The method adopts a fusion group constraint to reduce the variance of coefficients.•The modified discrete first-order method was used to obtain near optimal solutions.•Comparison studies show that the proposed method outperformed existing methods.

论文关键词：Multiple linear regression,Machine learning,Feature selection,Multicollinearity,Mixed-integer quadratic programming,Best subset selection

论文评审过程：Received 6 August 2017, Revised 12 February 2018, Accepted 13 May 2018, Available online 15 May 2018, Version of Record 19 May 2018.

论文官网地址：https://doi.org/10.1016/j.patcog.2018.05.013