SVFS

SVFS selects significant features from high-dimensional biological datasets using singular vectors and a dataset-derived signature matrix to reduce dimensionality while preserving informative signals for downstream analysis.

Key Features:

• Signature Matrix Utilization: SVFS employs a signature matrix derived from the dataset to cluster feature columns based on their correlations.
• Iterative Clustering and Reduction: SVFS iteratively uses the signature matrix to identify clusters of correlated features and discards features outside these clusters to reduce dimensionality.
• Cluster-Based Feature Selection: Within each identified cluster, SVFS partitions remaining features and selects the most important features from each cluster.
• Performance Efficiency: Experiments on synthetic and real-world datasets, including genomic data, report that SVFS outperforms existing state-of-the-art feature selection methods in accuracy, running time, and memory usage.

Scientific Applications:

• Dimensionality reduction in bioinformatics: SVFS reduces high-dimensional biological datasets while retaining informative features for analysis.
• Genomic data analysis: SVFS is applied to genomic datasets to identify informative genomic features for downstream study.
• Preprocessing for machine learning: SVFS provides feature sets that improve pattern recognition and classification performance of machine learning models.

Methodology:

SVFS computes a signature matrix to identify clusters of correlated features and discards features outside those clusters; the reduced dataset is then reprocessed via its signature matrix to partition clusters and select key features from each cluster.