Double-quantitative feature selection using bidirectional three-level dependency measurements in divergence-based fuzzy rough sets

Feature selection benefits machine learning and knowledge acquisition, it usually resorts to various intelligent methodologies. Fuzzy rough sets act as a powerful platform of processing, they have introduced divergence measures generate an effective method feature selection, called FS-DD. However, Algorithm FS-DD still has advancement space, because its underlying dependency degree with absoluteness lacks decision-categorical manifestations exhibits loose informatization. Within the framework divergence-based fuzzy (Div-FRSs), we implement bidirectional three-level measurements establish double-quantitative two novel approaches (i.e., Algorithms FS-AFS FS-RFS) are designed reconstruct improve current Based on lower-approximation matrices, first make in vertical horizontal directions, correspondingly absolute relative degrees. Then, degrees naturally induce significances, types uncertainty respectively exhibit granulation monotonicity non-monotonicity. Furthermore, significances utilized motivate algorithms, i.e., FS-RFS. Finally, measurement properties algorithms fully validated by table examples data experiments. This study systematically reveals hierarchical constructions quantitative characteristics Div-FRSs, effectively extract class-specific condensed information. For related interprets existing FS-DD, while new FS-RFS outperforms acquire better classification performances, experimentally verified.