Online Heterogeneous Feature Selection.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Real-world datasets often feature high-dimensional and heterogeneous attributes, complicating reliable feature selection and real-time analysis.
• Existing feature selection methods struggle with data heterogeneity, extreme dimensionality, and dynamic inter-feature relationships in data streams.

Purpose of the Study:

• To propose a new method for online heterogeneous feature selection (OHFS) that addresses the limitations of current approaches.
• To develop a robust framework for evaluating feature subsets in dynamic, heterogeneous data environments.

Main Methods:

• Introduced graph-unified adaptive decision boundary enhancement (GRADE) for OHFS.
• Developed an incremental graph-unified metric (IGUM) to unify heterogeneous feature relationships using graph structures.
• Proposed an adaptive density-guided neighborhood relation (ADNR) to assess feature subset classification capabilities in dynamic neighborhood regions.

Main Results:

• GRADE achieves a more concise feature subset while maintaining competitive classification accuracy.
• The proposed IGUM and ADNR effectively mitigate information loss and precisely delineate local decision boundaries.
• GRADE demonstrates parameter-free operation and superior efficiency compared to state-of-the-art methods.

Conclusions:

• GRADE offers an effective solution for online heterogeneous feature selection in complex, dynamic datasets.
• The method provides a reliable foundation for feature evaluation in evolving data streams.
• Experimental validation confirms the efficacy, efficiency, and conciseness of GRADE.