Feature selection is vital for improving high-dimensional data analysis by identifying a sub- set of representative and uncorrelated features. This paper presents an unsupervised feature selection algorithm based on subspace learning and adaptive graph structure (UFSAG). The UFSAG uses matrix factorization to preserve global data structure and incorporates local correlations into its objective func- tion. It also integrates sample similarity graph learning to maintain data geometry. Unlike prior methods, UFSAG employs adaptive local structure learning to reduce noise and enhance feature selection. By inducing row sparsity in the feature coefficient matrix using the ℓ2,1-norm, UFSAG identifies representa- tive features. Comparative experiments on six datasets show UFSAG’s superior clustering performance over twelve state-of-the-art methods.
