.png){kind=icon}
.png){kind=icon}
Return
Granular-ball based cross-granularity fuzzy knowledge collaborative feature selection
.png){kind=icon}
PRE
AI
B
H
L
F
W
DOI：10.1016/j.fss.2026.110032.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Granular-Ball Computing (GBC) is recognized for its robustness and efficiency in enhancing model performance with its coarse-granularity characteristic. In feature selection, GBC-enhanced models typically achieve significant improvements in the accuracy of subsequent classification tasks. However, existing methods usually remove samples during the granular-ball generation process rather than using the full dataset and subsequently rely solely on the centers of granular-balls for modeling. We provide the first quantitative evidence of this phenomenon using SHapley Additive exPlanations (SHAP) analysis, showing that such practices distort feature representations. To address this issue, we propose the Granular-Ball based Cross-Granularity Fuzzy Knowledge Collaboration (GCFKC) method for feature selection. At its core is the fuzzy rough sets model based on cross-granularity knowledge collaboration, which overcomes the issue of information loss in the traditional granular-ball generation process. By employing a structure-preserving incremental feature space updating strategy, we derive a feature significance evaluation metric. Results from comprehensive experiments conducted on public datasets demonstrate that our feature selection algorithm significantly outperforms existing methods in terms of both effectiveness and robustness.
Journal
IF:
2.7
Papers:
7.6K
Citations:
1.5W
Organization
Cited Papers