A hierarchical multilabel classification method based ON clustering relation
Hui Qi, XiaLin WANG, Xiaobo Qi, Xiaofang Mu, Kaige Duan, Chengpei Wu · IET conference proceedings. · 2022
In multilabel classification, the problems of a large number of classification calculations and easy destruction of label relations are very common. To solve these problems, a hierarchical multilabel classification method based on clustering relations is proposed by mining the possible dependencies between labels. First, the algorithm adopts a local strategy to cluster labels multiple times. Then, the clusters of labels with hierarchical relation are formed, and the implicit relationships hidden in these clusters are analyzed. On this basis, a multilabel clustered clustering tree is constructed to train the local model. Finally, the clustering tree is constructed as a random forest classification model using the ensemble idea. The experimental results show that the method in this paper has a good classification performance, especially on datasets with a large amount of data, which is at least percentage points higher than the existing algorithms.