S²FS: Spatially-Aware Separability-Driven Feature Selection in Fuzzy Decision Systems

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Mining

Background:

• Feature selection is vital for fuzzy decision systems (FDSs), enhancing performance and interpretability.
• Existing methods often neglect spatial data distribution and rely on limited distance metrics, hindering decision boundary clarity.
• The spatial arrangement of data points significantly influences class separability and decision boundary definition.

Purpose of the Study:

• To introduce a novel feature selection framework, spatially-aware separability-driven feature selection (S²FS), for fuzzy decision systems.
• To develop a spatially-aware separability criterion that integrates within-class compactness and between-class separation using spatial directional information.
• To improve the predictive performance and interpretability of FDSs by selecting more discriminative features.

Main Methods:

• Proposing the spatially-aware separability-driven feature selection (S²FS) framework for FDSs.
• Developing a novel criterion that combines scalar distances with spatial directional information for enhanced class structure characterization.
• Employing a forward greedy strategy for iterative selection of the most discriminative features.

Main Results:

• S²FS consistently outperformed ten state-of-the-art feature selection algorithms across 11 real-world datasets.
• The proposed method demonstrated superior performance in both classification accuracy and clustering.
• Feature visualizations confirmed the enhanced interpretability of features selected by S²FS.

Conclusions:

• S²FS offers a significant advancement in feature selection for FDSs by incorporating spatial awareness.
• The spatially-aware separability criterion effectively captures complex class structures, leading to improved model performance.
• S²FS provides a more interpretable and accurate approach to feature selection in fuzzy decision systems.