Incremental forward feature selection with application to microarray gene expression data.

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• Microarray gene expression datasets present challenges due to high dimensionality (many genes) and limited samples.
• Efficient feature selection is crucial for accurate classification and understanding biological data.

Purpose of the Study:

• To introduce and evaluate a novel feature selection scheme: incremental forward feature selection (IFFS).
• To assess the performance of IFFS against existing methods on complex gene expression datasets.

Main Methods:

• Developed an incremental forward feature selection algorithm inspired by incremental reduced support vector machines.
• Measured feature importance using the distance between a new feature vector and the column space of the current feature subset.
• Compared IFFS with the weight score approach and 1-norm support vector machine on acute leukemia and colon cancer datasets.

Main Results:

• The incremental forward feature selection and 1-norm support vector machine achieved slightly better classification results compared to the weight score approach.
• The proposed IFFS method effectively excludes highly linear correlated features, enhancing learning algorithm efficiency.
• Iterative removal and re-selection of features across four rounds revealed multiple distinct subsets capable of supporting classification tasks.

Conclusions:

• The incremental forward feature selection scheme is a viable and effective method for high-dimensional gene expression data.
• Distinct subsets of genes can provide sufficient information for classification, highlighting the redundancy and potential of various feature combinations.
• Further analysis suggests that remaining genes, not initially selected, may still contain valuable information for classification.