A block matrix incremental feature selection method based on fuzzy rough minimum classification error

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Inner-product correlation is key for feature selection in fuzzy rough set models, estimating minimum classification error.
• Existing methods using sample subsets limit accuracy in capturing global data structure.

Purpose of the Study:

• Propose a global-sample-oriented inner-product correlation criterion for enhanced feature evaluation.
• Develop static and incremental feature selection algorithms for improved performance and efficiency.

Main Methods:

• Constructing a continuous fuzzy membership structure over the entire data universe.
• Leveraging matrix computation for a static Minimum Classification Error-based Feature Selection (MCEFS) algorithm.
• Implementing a block-wise updating mechanism for dynamic data environments, leading to Block Matrix-based MCEFS (BM-MCEFS).

Main Results:

• The global-sample criterion enhances theoretical soundness and practical consistency.
• The static MCEFS algorithm demonstrates effectiveness and feasibility.
• BM-MCEFS shows superior computational efficiency and numerical stability on benchmark datasets.

Conclusions:

• The proposed global-sample approach advances inner-product-based feature selection.
• BM-MCEFS offers an efficient solution for feature selection in dynamic environments.
• The study validates the effectiveness and superior performance of the developed algorithms.