Supervised, Unsupervised, and Semi-Supervised Feature Selection: A Review on Gene Selection.

Area of Science:

• Bioinformatics
• Data Mining
• Machine Learning

Background:

• High-dimensional data, such as gene expression microarrays, present challenges for machine learning algorithms due to a large number of features and small sample sizes.
• Feature selection and dimensionality reduction are essential for improving the performance of data mining tasks and cancer classification.
• Numerous gene selection methods exist, broadly categorized into supervised, unsupervised, and semi-supervised approaches.

Purpose of the Study:

• To present a basic taxonomy of feature selection methods.
• To review state-of-the-art gene selection techniques.
• To compare the performance of different gene selection categories on cancer classification.

Main Methods:

• Literature review and categorization of gene selection methods into supervised, unsupervised, and semi-supervised.
• Experimental comparison of these methods on five representative gene expression datasets.

Main Results:

• Gene selection is fundamental for processing high-dimensional data like gene expression microarrays.
• Unsupervised and semi-supervised gene selection methods achieve classification accuracy comparable to supervised methods.
• The study provides a comparative analysis of gene selection techniques on real-world datasets.

Conclusions:

• Unsupervised and semi-supervised feature selection offer competitive performance for cancer classification.
• These methods are viable alternatives to supervised approaches for analyzing gene expression data.
• The findings support the utility of diverse gene selection strategies in bioinformatics.