Least squares twin bounded support vector machines based on L1-norm distance metric for classification

作者：

Highlights：

• We have enhanced TBSVM to LSTBSVM in least squares sense, while in LSTBSVM the distance is measured by L1-norm.

• L1-LSTBSVM has more robustness to outliers, can lower the computational costs and improve the classification performance.

• We design a valid iterative algorithm to solve the L1-norm optimal problems, which is an important theoretical contribution.

• The method which we proposed can be conveniently extended to solve other improved methods of TWSVM.

摘要

•We have enhanced TBSVM to LSTBSVM in least squares sense, while in LSTBSVM the distance is measured by L1-norm.•L1-LSTBSVM has more robustness to outliers, can lower the computational costs and improve the classification performance.•We design a valid iterative algorithm to solve the L1-norm optimal problems, which is an important theoretical contribution.•The method which we proposed can be conveniently extended to solve other improved methods of TWSVM.

论文关键词：L1-LSTBSVM,TBSVM,L1-norm distance,Outliers

论文评审过程：Received 26 September 2016, Revised 19 September 2017, Accepted 23 September 2017, Available online 25 September 2017, Version of Record 5 October 2017.

论文官网地址：https://doi.org/10.1016/j.patcog.2017.09.035