Research On Filtering Multi-label Feature Selection Algorithm Based On Swarm Intelligence Optimization

Classification is an important and common problem in the field of machine learning.According to the number of labels corresponding to each sample,it can be divided into the following two categories: the first category is single label classification,and the traditional single label classification processes the task of labeling each sample with a class label.The second is multi label classification,which deals with the task of labeling each sample with multiple class labels.In the real world,the data sets of multi label classification usually have high-dimensional features and the features contained in these high-dimensional features are usually uncorrelated and redundant,which improves the complexity of classification and affects the performance of classifier,so it is a very difficult task to build the classification model from the real world data sets.As a very important data preprocessing step in the field of machine learning,feature selection plays an important role in multi label classification.This paper proposes two multi-label feature selection methods based on a filtered multi-label feature selection algorithm and swarm intelligence optimization:(1)Multi-label feature selection algorithm based on normalized mutual information and particle swarm algorithm with binary discrete strategy(NMI-BDPSO);(2)Multi-label feature selection algorithm based on correlation and mutation binary bat algorithm(CFS-MBBA).For the multi-label feature selection algorithm NMI-BDPSO,by using normalized mutual information to measure the relationship between features and characteristics,features and labels,and labels,a criterion function based on normalized mutual information was constructed.The particle swarm algorithm with discrete strategy uses the criterion function as a fitness function to select the optimal feature subset of a given size.In the experimental stage,the performance of the proposed algorithm NMI-BDPSO and the other three multi-label feature selection algorithms on four multi-label data sets is compared on 10 multi-label classification evaluation indicators.The final results show that the proposed algorithm NMI-BDPSO has better performance.For the multi-label feature selection algorithm CFS-MBBA,the function of the correlation criterion is used in combination with the similarity measurement method of Pearson correlation coefficient to measure the correlation between features and labels,and the redundancy between features.The fitness function of the mutated binary bat algorithm is a function of the correlation criterion,which is used to search the feature subset and select the optimal feature subset.In the experimental phase,the proposed algorithm CFS-MBBA was compared with three other comparison algorithms on ten multi-label classification evaluation indexes on seven multi-label data sets.The final results show that the proposed algorithm CFS-MBBA classification is better.