Abstract: High computational complexity limits the applications of the Bilateral-Weighted Fuzzy Support Vector Machine (BW-FSVM) model in practical classification problems. In this paper, the Sequential Minimal Optimization (SMO) algorithm,which firstly decomposed the overall Quadratic Program (QP) problem into the smallest possible QP sub-problems and then solved these QP sub-problems analytically, was proposed to reduce the computational complexity of the BW-FSVM model. A set of experiments were conducted on three real world benchmarking datasets and two artificial datasets to test the performance of the SMO algorithm. The results indicate that compared with the traditional interior point algorithm, the SMO algorithm can reduce significantly the computational complexity of the BW-FSVM model without influencing the testing accuracy, and makes it possible for the BW-FSVM model to be applied to practical classification problems with outliers or noises.

Key words: Sequential Minimal Optimization (SMO), Bilateral-weighted fuzzy support vector machine, Support Vector Machine (SVM), Fuzzy Support Vector Machine (FSVM)

摘要： 高的计算复杂度限制了双边加权模糊支持向量机在实际分类问题中的应用。为了降低计算复杂度，提出了应用序贯最小优化算法（SMO）解该模型，该模型首先将整个二次规划问题分解成一系列规模为2的二次规划子问题，然后求解这些二次规划子问题。为了测试SMO算法的性能，在三个真实数据集和两个人工数据集上进行了数值实验。结果表明：与传统的内点算法相比，在不损失测试精度的情况下，SMO算法明显地降低了模型的计算复杂度，使其在实际中的应用成为可能。

关键词: 序贯最小优化, 双边加权模糊支持向量机, 支持向量机, 模糊支持向量