基于分类器链的多标签分类算法

Multi-Label Classification Algorithm Based on Classifier Chain

为解决分类器链中错误信息传递和信息冗余的问题,提出了一种标签选择有序分类器链算法(LS-OCC)。首先通过对标签进行排序形成有序分类器链以减小错误信息在链中的传递,然后通过计算标签间的相关性对标签进行选择,在充分考虑标签间相关性的同时又降低了分类器属性空间的信息冗余。在实验中采用Mulan中的8个多标签基准数据集,相比于分类器链(CC)、有序分类器链(OCC)、贝叶斯链分类器(BCC)等算法,LS-OCC算法在准确率、汉明损失和Macro-F1上都取得了更好的分类效果。

In order to solve the problem of error information transmission and information redundancy in the classifier chain algorithm, a label selection ordered classifier chain algorithm(LS-OCC) is proposed. First, the labels were sorted to form an ordered classifier chain to reduce the transmission of error information in the chain. And then the labels were select by calculating the correlation between the labels, which not only considers the correlation between the labels, but also reduces the information redundancy in the attribute space of the classifier. The 8 multi-label benchmark datasets in Mulan were used in the experiment. Compared with Classifier Chain(CC), Ordered Classifier Chain(OCC), Bayesian chain classifier(BCC) and other algorithms, the LS-OCC algorithm achieves better classification performance in accuracy, Hamming loss and Macro-F1.