摘要： 最小二乘支持向量机在提高求解效率的同时，会丧失解的稀疏性，导致其在预测新样本时速度较慢。为此，提出一种稀疏化最小二乘支持向量机分类算法。在特征空间中寻找近似线性无关向量组，构造分类判别函数的稀疏表示，相应的最小二乘支持向量机优化问题可以通过线性方程组求解，从而得到最优判别函数。实验结果表明，该算法在不损失分类精度的前提下，能够获得比最小二乘支持向量机更快的预测速度。

关键词: 支持向量机, 最小二乘, 稀疏化, 分类, 特征空间, 二次规划

Abstract: Least Squares Support Vector Machine(LSSVM) increases the algorithm’s efficiency, but the sparsity of the solution is lost, which results in the low prediction speed of new samples. A new sparsified LSSVM classification algorithm is proposed. The approximate linear independent vector set in feature space is first searched and thus the sparse discriminative function can be constructed. The corresponding optimization problem of LSSVM can then be solved through the linear equation set. So the optimal discriminative function is achieved. Experiments show that the prediction speed of the proposed algorithm is faster than that of LSSVM without the loss of classification accuracy.

Key words: Support Vector Machine(SVM), least squares, sparsification, classification, feature space, quadratic programming

中图分类号: 

• TP311.13