基于LS-SVM的多标签分类算法

A Multi-Label Classification Algorithm Based on LS-SVM

多标签分类是指部分样本同时归属多个类别.基于数据分解的算法因训练速度快、性能良好而得到广泛的应用.本文采用一对一分解策略,将k标签数据集分解为k(k-1)/2个两类单标签和两类双标签的数据子集.对每一训练子集统一用LS-SVM模型建立子分类器,当出现双标签样本时将其函数值设为0,并确定适当的分类阈值.对情感、景象和酵母数据集的实验结果表明,本文算法的某些性能指标优于现有一些常用的多标签分类方法.

A multi-label classification problem lies in that its samples may belong to multiple classes.Data decomposition algorithms are widely used because of its good performance.One versus one decomposition strategy is adopted in this paper,and this strategy decomposes a multi-label problem into several binary class single label or binary class double label classification sub-problems which can be solved independently.For each sub-problem,we build a sub-classifier using LS-SVM model and set the function value zero when the sample is double label,then determine a proper threshold.Experimental results show that our performance is superior to several existent multi-label classification algorithms with some evaluation criteria on three benchmark datasets Yeast,Scene and Emotion.