Patch Uniform Fusion Transformer with circle-inspired optimized walrus-based feature selection for high-dimensional...

Area of Science:

• Machine Learning
• Data Science
• Bioinformatics

Background:

• High-dimensional data analysis faces performance limitations with traditional and hybrid feature selection methods.
• Existing approaches struggle with computational intensity and feature reduction, impacting accuracy and speed.

Purpose of the Study:

• To introduce an efficient hybrid feature selection approach, Improved Circle-Inspired Walrus Optimization (ICIWO), for very high-dimensional datasets.
• To address the computational and accuracy limitations of current feature selection techniques.

Main Methods:

• Proposed Improved Circle-Inspired Walrus Optimization (ICIWO) for hybrid feature selection.
• Utilized Fully Spiking Variational Autoencoder (FSVA) with spiking neural networks for data compression and dimensionality reduction.

Main Results:

• ICIWO achieved high performance on the ALL-AML dataset with 98% accuracy, 0.9913 AUC, and 99.28% F1-score.
• Demonstrated strong results on GLI-85 (97% accuracy, 0.9857 AUC) and CLL-SUB-111 (95% accuracy, 0.9753 AUC) datasets.
• ICIWO showed a fitness value of 0.0607 on ALL-AML and 0.0727 on GLI-85.

Conclusions:

• ICIWO offers a robust and cost-effective solution for high-dimensional data analysis.
• The method overcomes limitations of existing techniques, enabling improved pattern recognition and insights.
• ICIWO demonstrates significant potential for diverse datasets requiring efficient feature selection.