Stable Iterative Variable Selection.

Area of Science:

• Biomedical data analysis
• Bioinformatics
• Machine learning in healthcare

Background:

• High-throughput omics and clinical datasets present challenges due to a large number of features.
• Effective feature selection is crucial for identifying biomarkers and building robust predictive models.
• Reducing feature space improves model interpretability and convergence.

Purpose of the Study:

• To introduce Stable Iterative Variable Selection (SIVS), a novel feature selection method.
• To evaluate SIVS performance on omics and clinical data.
• To compare SIVS with existing feature selection techniques.

Main Methods:

• Stable Iterative Variable Selection (SIVS) algorithm.
• Performance comparison using feature set size and model performance metrics.
• Evaluation on omics and clinical datasets.

Main Results:

• SIVS selected feature sets that were, on average, 41% smaller than those from Least Absolute Shrinkage and Selection Operator regression.
• The reduced feature space from SIVS did not negatively impact model performance.
• Similar improvements in feature reduction were observed when compared to Boruta and caret RFE.

Conclusions:

• SIVS is a robust feature selection method for high-dimensional biomedical data.
• The method effectively reduces feature space while maintaining or improving model performance.
• SIVS offers a valuable tool for biomarker discovery and predictive modeling in research and industry.