Research On Embedded Multi-label Feature Selection Algorithm

With the rapid development of science and technology,the amount of data that needs to be processed in the classification problem has increased dramatically,not only the traditional single-label data,but also the multi-label data in the real world.The same as the single-label classification problem,the samples in the multi-label classification problem usually also contain multiple features,among which redundant and unrelated features will affect the performance of the classifier.To solve this problem,multi-label feature selection has become a research hotspot in the field of machine learning in recent years.At present,the existing embedded multi-label feature selection methods are mostly sparse regularization feature selection methods using least squares regression as the objective function.These methods rarely consider the geometry of the feature manifold.Since the correntropy has very good robustness when dealing with outliers,it can better deal with nonlinear and non-gaussian data,and can eliminate the harmful effects of outliers.Therefore,the correntropy is used as the objective function of multi-label feature selection.Combined with feature manifold learning,a sparse regularized multi-label feature selection algorithm based on correntropy and feature manifold learning is proposed and a algorithm based on K-Support norm and manifold learning multi-label feature selection is proposed in this paper.(1)In order to better reflect the importance of label information,by changing the distance between different categories of labels in multi-label datasets,and then combining l2,1 norm,a least squares multi-label feature selection with label information model and algorithm are proposed.Finally,the efficiency of the algorithm is proved by experiments.(2)Based on the idea of correntropy,combing with l2,1 norm and feature manifold learning to construct multi-label feature selection regression model,a sparse regularized multi-label feature selection model and algorithm based on correntropy and manifold learning is proposed and the effectiveness of the given algorithm is verified through experiments.(3)In order to better consider the correlation between data,a multi-label feature selection model and algorithm based on K-Support norm and manifold learning is proposed.And then the effectiveness of the given algorithm is verified by experiments.