Adaptive fuzzy cluster-guided simple, fast, and efficient feature selection for high-dimensional and highly...

Area of Science:

• Bioinformatics
• Machine Learning
• Data Mining

Background:

• High-dimensional, imbalanced microarray data present challenges like feature redundancy and class overlap.
• These issues bias learning algorithms towards the majority class, hindering accurate classification.

Purpose of the Study:

• To propose the Adaptive Fuzzy Cluster-Guided Simple, Fast, and Efficient (AFCG-SFE) feature selection model.
• To address feature redundancy and class overlap in imbalanced microarray data for improved classification.

Main Methods:

• AFCG-SFE utilizes two-stage fuzzy feature clustering and mutual information for feature selection.
• It incorporates an imbalance-aware penalty-reward fitness function optimizing F-measure, G-mean, and AUC.
• A complexity-driven minimum subset size is enforced using feature separability (F1) and class overlap (N2).

Main Results:

• AFCG-SFE achieved top-tier classification performance across 20 benchmark datasets.
• The model significantly reduced feature subsets (feature redundancy reduction > 99%) and class overlap (N2).
• It demonstrated the lowest train-test Root Mean Square Error (RMSE) compared to baselines.

Conclusions:

• The AFCG-SFE model offers a robust solution for feature selection in high-dimensional, imbalanced microarray data.
• It effectively balances feature discrimination, redundancy reduction, and minority class sensitivity.
• AFCG-SFE outperforms existing methods in classification accuracy and feature subset reduction.