Missing Multi-label Learning Of Imbalanced With Label Reconstruction

Multi-label learning can effectively adapt to multi-semantic problems in the real world,which has been a hot research interest in machine learning.With the increase of data volume and complexity,researchers introduced label correlation to improve the accuracy of approaches for multi-label learning.However,only considering the correlation of labels may reduce the robustness of the approach.In addition,the imbalance of labeling and label diversity in multi-labellearningdatasets will directly affect multi-label learning approach performance.Accordingly,classification and modeling against imbalanced data is necessary.Meanwhile,labeling massive data is difficult andsome labels are inevitably missing.Therefore,how to perform label completion and reduce the missing label interference is vital for improving the accuracy of multi-label classification.Based on the above issues,the main contents are as follows:1)The existing multi-label learning algorithms only consider the correlation among labels and ignore the influence of the feature space.According to the idea of the firefly method,a Multi-labelLazy Learning Approach based on FireFly method(FF-MLLA)is proposed,which?reconstruct?label?space?by?combining?feature?information?and?label?information.Firstly,Minkowski distance is used to measure the similarity between samples to find the nearest neighbor point.Secondly,the label count vector is improved by combining the nearest neighbor point and firefly method.Finally,the singular value decomposition and the kernel extreme learning machine are used to classify the linear classifier.The approach considers both the label information and similarity information to improve the robustness of the approach.2)Different from label imbalances in single-label learning,internal imbalance and imbalance between labels are common in multi-label learning datasets.Current methods mostly focus on combining sampling techniques with cost-sensitive learning and incorporating label correlation to improve the performance of the classifier,but generally they do not consider label loss caused by label cost.In fact,labeling unknown instances is often affected by the threshold of the discriminant function,especially for the label types near the threshold.Based on our previous research,we believe that information such as data distribution density and label density can be integrated into the label correlation,and that the classification margin can be expanded to effectively solve the labeling quality of labels near the threshold.Therefore,we propose a missing Multi-label learning with Non-Equilibrium based on Classification Margin(MNECM),which aims at completing the missing labels3)With the increase of data volume,the common absence of some label data results in the incompleteness of the label space,and leads to difficulties in measuring the label correlation.Many multi-label learning algorithms focus on label correlation for missing label recovery,but ignore the instance information.Therefore,we use the attention mechanism to jointly exploit the label and the instance information in order to improve the label quality and recovery.Based on this,we propose a Global and Local Multi-label learning withAttention mechanism for Missing labels(GLMAM).In this thesis,the firefly method is used to reconstruct label space via combining label and feature information,we expand the classification margin to solve the label imbalance.The non-equilibrium method and attention mechanism are used to perform label completion.Through the analysis of the experimental results of multiple benchmark multi-label datasets,it is shown that the proposed method has certain advantages over other state-of-the-art multi-label learning algorithms.The statistical hypothesis testing and stability analysis are used to further illustrate the rationality and effectiveness of the proposed approach.