隶属度阈值函数极限学习机的多标签学习

Multi-label Learning based on the Similar Membership Threshold Function Extreme Learning Machine

在传统的多标签学习算法中,对标签的阈值函数通常选用Sign函数进行计算,忽略了标签的差异性,直接影响到多标签分类结果.虽然目前对多标签学习的研究较多,但关于Sign函数及其阈值的确定研究不多.基于此,提出隶属度阈值函数核极限学习机(SMF-ML-RKELM)算法:首先利用核极限学习机训练出结果,再利用训练结果与训练集标签进行比对,计算出每一个标签的隶属度阈值函数,最后使用测试集进行预测.该算法有效结合了阈值自适应的误差减小策略及其核极限学习机学习的优点,实验结果表明该算法与其他几种分类算法相比,具有较好的分类精度.

In the traditional multi-label learning algorithm,the labeled threshold function is usually calculated by the Sign function,ignoring the differences of the labels. However,the labels in each data concentration have their own features. The Sign function is used to determine the result which directly affects the results of multi-label classification.Although there have been many researches on multi-label learning,there are few explorations of changes to the threshold function. Based on this, this paper proposes the similar membership threshold function and Extreme Learning Machine(SMF-ML-RKELM) algorithm. Firstly,the training results are trained by using the kernel extreme learning machine,and then the training results are compared with the training set labels to calculate the class membership threshold function of each label. Finally,the test set is used to predict the class membership threshold function of each label. The algorithm effectively combines the advantages of threshold adaptive error reduction strategy and kernel extreme learning machine.The experimental results show that the algorithm has better classification accuracy than other classification algorithms. The experimental results show that the algorithm has better classification accuracy than other classification algorithms.