Multi-label Learning Algorithm Based On Bayesian Network Model

With the rapid development of information technology,a large amount of multi-label data emerge in real life.In recent years,the multi-label learning has become a hot issue.At present,researchers have proposed a variety of multi-label learning methods.Most of the methods have been divided into two types,the one is the problem transformation method whose main idea is to deal the multi-label training data and transform multi-label algorithm into other single label algorithm.The other is algorithm adaptation method,whose main idea is to improve the common supervised learning algorithm.Because label dependence are not accounted in current proposed algorithms and bayesian network can find variables dependence through data analysis.Bayesian network is adopt to study the multi-label classification problem in the paper.Based on the label dependence and ML-k NN algorithm,the multi-label classification algorithms,such as the label dependence classification algorithm and multi-label classification algorithm by constructing k-nearest neighbor data,are proposed.The main work is as follow:(1)Multi-label data sets discretization algorithm was proposed by introducing mlaim.Multi-label classification algorithm based on label dependence was proposed by the structure learning of Bayesian networks on the label set,and the efficiency of the algorithm is proved by some numerical experiments.(2)By the idea of k-nearest neighbor classification,a new data set is constructed by using the labels of each data k-nearest neighbor,and the linear regression model and the Logistic regression model are established under the new data set.Moreover,the multi-label classification algorithms by constructing k-nearest neighbors data are proposed,and the efficiency of the algorithms are verified by some numerical experiments.(3)In order to further use the information of original data set,Markov boundary of each label has been considered,and regression model by using Markov boundary as attributes for new data sets has been established.Moreover,multi-label classification algorithm of considering Markov boundary was proposed and the efficiency of the algorithm is proved by some numerical experiments.