Robust Semi-Supervised Feature Selection With Multi-Granularity Zentropy Modeling.

Area of Science:

• Machine Learning
• Pattern Recognition
• Data Science

Background:

• High-dimensional and weakly supervised (HiDWS) data pose challenges for traditional machine learning.
• Existing semi-supervised feature selection methods lack robustness due to unreliable unlabeled data and uncertain modeling.

Purpose of the Study:

• To propose a novel multi-granularity zentropy modeling (Ze-MGM) framework for highly-accurate and robust semi-supervised feature selection.
• To address the limitations of existing methods in handling HiDWS data.

Main Methods:

• Introduced a strategic soft label ($S2-$Label) learning method integrating object proximity and classification certainty.
• Constructed a multi-granularity knowledge space and zentropy uncertainty measure analyzing label-decision-class hierarchies.
• Defined two multi-granularity significance measures for feature evaluation and selection.

Main Results:

• The proposed Ze-MGM framework effectively captures information granularity in HiDWS data.
• Ze-MGM reduces uncertainty between features and labels by selecting compatible instances.
• Achieved superior generalization performance and robustness compared to state-of-the-art methods in extensive experiments.

Conclusions:

• Ze-MGM offers a robust and accurate solution for semi-supervised feature selection in HiDWS data.
• The framework's model-agnostic nature and ability to capture multi-granularity information contribute to its effectiveness.