Dimensionality reduction using singular vectors.

Area of Science:

• Bioinformatics
• Machine Learning
• Pattern Recognition

Background:

• High-dimensional datasets in bioinformatics present challenges for feature selection.
• Identifying relevant features is crucial for accurate pattern recognition and machine learning model performance.

Purpose of the Study:

• To introduce a new feature selection method, Singular-Vectors Feature Selection (SVFS).
• To demonstrate SVFS's effectiveness in reducing dimensionality and selecting optimal features from high-dimensional datasets.

Main Methods:

• SVFS utilizes a signature matrix to partition feature columns into clusters.
• It identifies and retains clusters relevant to the class label, discarding irrelevant features.
• The method iteratively refines feature selection by clustering remaining features.

Main Results:

• SVFS effectively reduces dataset size by discarding irrelevant features.
• Experiments show SVFS outperforms state-of-the-art methods in accuracy, speed, and memory efficiency.
• The method performs well on both synthetic and real-world benchmark/genomic datasets.

Conclusions:

• SVFS is a superior feature selection technique for high-dimensional bioinformatics data.
• The method offers significant improvements in performance metrics compared to existing approaches.
• A Python implementation is publicly available for use and further research.