Hierarchical Multi-label Integrated Chain Evolutionary Hypernetwork

Generally,the biggest difference between multi-label learning and single-label learning lies in that labels of multi-label data are organized in a label vector,and labels often show complex correlations.Hierarchical multi-label classification is a special type of multi-label classification.This thesis mainly studies the hierarchical multi-label classification problem where a sample can be associated with multiple class labels residing on a hierarchy.The main challenges of hierarchical multi-label classification lie in the following two aspects: 1)the predictions must meet the label hierarchy constraints;2)labels at lower levels usually have few related samples,leading to the imbalance problem among the labels.In order to deal with these challenges,following researches are conducted:1.Although the traditional multi-label hypernetwork can cope with multi-label classification problems effectively by utilizing correlations among labels,it fails to model the hierarchical structure of labels in hierarchical multi-label classification.Thus the traditional multi-label hypernetwork cannot be directly used to solve hierarchical multi-label classification problems.In order to handle hierarchical multi-label classification problems,an incremental multi-label hypernetwork for hierarchical multi-label classification(HMC-IMLHN)is proposed.By organizing hyperedges of hypernetwork into corresponding hierarchy,the prediction of hypernetwork can meet label hierarchy automatically.2.In hierarchical multi-label data,labels have a very complex structure.In addition to the difference between single paths and multi-paths,there also exist other very complicated situations.In order to improve the robustness and accuracy of the hierarchical multi-label classification,a hierarchical multi-label integrated chain evolutionary hypernetwork(HMC-ICMLHN)is proposed.HMC-ICMLHN takes both single-path and multi-paths into consideration,reconstructs data by utilizing Bagging algorithm on clustered data,and classifies data samples using classifier chains.In order to verify the hierarchical multi label classification method proposed in this paper is effective,the experimental part is verified by open hierarchical multi label dataset.The experimental results show that the proposed methods can effectively deal with hierarchical multi-label classification problems.