Multi-label Learning Based On Label Weight And Weighted Kernel Extreme Learning Machine

Multi-label learning is one of the research emphases in machine learning and data mining.The purpose of multi-label learning is to predict the unknown samples accurately by analyzing the existing multi-label data.In most multi-label data sets,there are many redundant features in the feature space of the description sample.Redundancy feature not only affects the accuracy of classification,but also increases the complexity of calculation.Feature selection is one of the effective methods to solve this problem.However,in the practical classification problem,the class imbalance problem is also one of the research difficulties of machine learning.This paper makes a full investigation of relevant references,summarizes the research status,and analyzes the advantages and disadvantages of the existing methods.On this basis,the discriminability of value of the label to the sample can be mined,and give the weight value of the label.A solution to the class imbalance problem is proposed.The main research work of this paper is as follows:(1)Mining out the discriminability of label pairs,and according to the weight of label pairs,a multi-label feature selection based on kernel function and label weight(KF-LW)are proposed.First of all,the number of samples with different labels was counted.If the number of samples attached to a label is significantly higher than that of samples containing other labels,it indicates that the weight of the label is larger,and the weight of the label is assigned according to the information of the label space.Then,the kernel function is used to map the original feature space to the high dimensional space,so that the feature is separable.Finally,the feature subset is selected according to the correlation between the information entropy measurement feature and the label space.(2)According to class imbalance problem,based on the weighted nuclear extreme learning machine,and its application to the multi-label classification problem,a multi-label learning algorithm-based on weighted kernel extreme learning machine is proposed(ML-WKELM),the algorithm through the calculation the number of each sample and statistical average labels,weight matrix is calculated for each sample,The classifier assigns more weight to a few class samples and less weight to most class samples,so as to solve the unbalanced problem of multi-label class imbalance problem and improve the classification accuracy.