Semi-supervised multi-label feature selection with local logic information preserved

In reality, like single-label data, multi-label data sets have the problem that only some labels. This is an excellent challenge for feature selection. paper combines logistic regression model with graph regularization and sparse to form a joint framework (SMLFS) semi-supervised First of all, used explore geometry structure feature, obtain better coefficient matrix, which reflects importance feature. Second, label extract available information, constrain so matrix can fit information. Third, $$L_{2,p}$$ -norm $$0<p\le 1$$ constraint ensure sparsity it more convenient distinguish features. addition, iterative updating algorithm convergence designed proved solve above problems. Finally, proposed method validated on eight classic sets, experimental results show effectiveness algorithm.