基于两步特征加权的模糊支持向量机算法

Fuzzy support vector machine algorithm based on two-step feature weighting

提出一种基于两步特征加权的模糊支持向量机算法.首先,利用信息增益算法获取样本的特征权重.然后,计算最大权重的特征与其他特征间的斯皮尔曼相关系数,并将二者相乘后再与原有的特征权重相加,得到新的特征权重,减少弱相关和不相关特征对分类造成的影响.最后,在设计样本模糊隶属度时,不仅考虑样本与类中心的距离,还引入了样本间的亲和度,并将二者进行融合,以此减弱样本分布不均对分类精度的影响.在UCI数据集上的实验表明,与现有流行的几种模糊支持向量机算法相比,所提算法在准确率和F1值上得到了提升.

A fuzzy support vector machine algorithm based on two-step feature weighting is proposed. Firstly, the information gain algorithm is used to obtain the feature weights of the samples. Then, the Spearman correlation coefficients between the feature with the maximum weight and other features are calculated, and the corresponding Spearman correlation coefficients are multiplied by the maximum feature weight. Then the results are added with the original feature weights to get the new feature weights, so as to reduce the impact of weakly correlated features and irrelevant features on classification. Finally, when designing the fuzzy membership of samples, not only the distance between samples and class center is considered, but also the affinity between samples is introduced. And the distance and the affinity are fused so as to reduce the influence of uneven distribution of samples on classification accuracy. Experiments on UCI dataset show that compared with several popular fuzzy support vector machine algorithms, the proposed algorithm is improved in accuracy and F 1 value.