基于L1-范数距离的最小二乘对支持向量机

L1-norm Distance Based Least Squares Twin Support Vector Machine

最小二乘对支持向量机(LSTSVM)是一种有效的分类技术。然而,该方法需计算点到平面的平方L2-范数距离,从而易受野值或噪声的影响。为了缓解此问题,提出了一种有效的鲁棒LSTSVM方法,即基于L1-范数距离的LSTSVM(LSTSVM_(L1D))。该方法由于使用L1范数作为距离度量,因此不易受到野值或噪声数据的影响。此外,设计了一种有效的迭代算法,旨在求解目标问题,并从理论上证明了其收敛性。在人工数据集和UCI数据集上验证了LSTSVM_(L1D)的有效性。

Recently,LSTSVM,as an efficient classification algorithm,was proposed.However,this algorithm computes squared L2-norm distances from planes to points,such that it is easily affected by outliers or noisy data.In order to avoid this problem,this paper presented an efficient L1-norm distance based robust LSTSVM method,termed as LSTSVM L1D.LSTSVM L1D computes L1-norm distances from planes to points and is not sensitive to outliers and noise.Besides,this paper designed an efficient iterative algorithm to solve the resulted objective,and proved its convergence.Experiments on artificial dataset and UCI dataset indicate the effectiveness of the proposed LSTSVM L1D.