Research On Multi-label Classification Algorithm Based On Label Correlation And Three-layer Bp Neural Network

Nowadays,multi-label classification is already a hot topic for discussion at all major machine learning conferences.In the traditional classification problem,a sample can only correspond to a label.But in a multi-label classification problem,a sample often corresponds to a set of labels,which usually contain two or more labels.So the traditional method has been unable to meet the needs of multi-label classification.Multi-label classification problems often occur in real life.Over the past decade,researchers at home and abroad have explored this problem and produced some related results,and published a variety of multi-label classification algorithms.One of the Multi-label classification algorithms using three-layer BP neural network shows good performance,and the application field is wide and can be applied to a variety of different classification databases.But this algorithm often has some shortcomings,such as algorithm performance Low,the algorithm runs longer.The reason is largely due to the following two parts:on the one hand,some algorithms do not take into account the correlation between labels,on the other hand some of the algorithms can not completely or correctly describe the correlation between the labels.So there are still many aspects of this algorithm can be studied.In this paper,we start from the current research status of multi-label classification,and then describe the relevant issues,and discuss the challenges to solve the problem.This paper summarizes the performance measurement and evaluation methods of the current multi-label classification algorithm.The basic knowledge of the three-layer BP neural network is reviewed in detail,including the network structure,the learning process,the training process,the selection of some parameters in the network,and some methods of preprocessing the input data.Then,this paper summarizes four algorithms which use three-layer BP neural network for multi-label classification,and prove that there are some advantages in the algorithm to consider the relationship between labels.Finally,a multi-label classification algorithm based on label correlation and three-layer BP neural network is proposed on the basis of predecessors.In this algorithm,the relationship between the relevant label and the irrelevant label in the label set,the relationship between the relevant labels and the relation between the irrelevant labels are fully considered.And the experimental results show that the algorithm show better performance compared to other algorithms in this paper.