面向不均衡分类的隶属度加权模糊支持向量机

针对不均衡分类问题，提出了一种基于隶属度加权的模糊支持向量机模型。使用传统支持向量机对样本进行训练，并通过样本点与所得分类超平面之间的距离构造模糊隶属度，这不仅能够消除噪点和野值点的影响，而且可以在一定程度上约减样本；利用正负类的平均隶属度和样本数量求得平衡调节因子，消除数据不平衡时造成的分类超平面的偏移现象；通过实验结果验证了该算法的可行性和有效性。实验结果表明，该算法能有效提高分类精度，特别是对不平衡数据效果更加明显，在训练速度和分类性能上比传统支持向量机和模糊支持向量机有进一步的提升。

Weighted fuzzy support vector machine faced on fuzzy membership of imbalanced classification

In view of the classification of imbalanced data set, a weighted fuzzy support vector machine is proposed, making use of the balanced adjustment factor and the fuzzy membership based on the features of samples. Firstly, it trains the classification hyperplane by traditional support vector machine and gets the fuzzy membership of every sample to be considered as the contribution rate of every sample to eliminate the error caused by noises and outliers and subtract the number of samples in a certain extent. Subsequently, it computes the balanced adjustment factor to alleviate the migration of hyperplane. Ultimately, experiments on a number of real-world data sets even including the data sets are imbalanced show that the proposed weighted fuzzy support vector machine algorithm is scalable and outperforms the existing fuzzy support vector machine as well as the typical support vector machine counterparts.