CGUFS: A clustering-guided unsupervised feature selection algorithm for gene expression data.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Gene expression data is high-dimensional with many irrelevant features, posing challenges for analysis.
• Existing unsupervised feature selection methods often ignore feature redundancy and struggle with determining the optimal number of features.

Purpose of the Study:

• To propose a clustering-guided unsupervised feature selection (CGUFS) algorithm for gene expression data.
• To address limitations in existing algorithms regarding feature redundancy and optimal feature subset selection.

Main Methods:

• Developed an adaptive k-value strategy for automatic cluster number determination.
• Implemented a feature grouping strategy to manage highly redundant features.
• Introduced an adaptive filtering strategy to select optimal feature combinations.

Main Results:

• The CGUFS algorithm achieved average accuracy (ACC) of 74.37% and Matthews Correlation Coefficient (MCC) of 63.84% with the C4.5 classifier.
• Significantly superior performance was observed with the Adaboost classifier compared to existing algorithms.
• Statistical experiments confirmed significant differences in performance between CGUFS and existing methods.

Conclusions:

• The proposed CGUFS algorithm effectively selects optimal features from high-dimensional gene expression data.
• CGUFS demonstrates superior classification performance and addresses key limitations of current unsupervised feature selection techniques.