Related Experiment Video
Updated: Aug 13, 2026
07:35
Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances
Published on: October 11, 2018
7.8K
RIFS2D: A two-dimensional version of a randomly restarted incremental feature selection algorithm with an application
Sida Gao1, Puli Wang1, Yuming Feng1
1College of Computer Science and Technology, Jilin University, Changchun, Jilin, 130012, China; Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education, Jilin University, Changchun, Jilin, 130012, China.
Computers in Biology and Medicine
|April 30, 2021
Summary
Biomedical researchers can improve prediction models by using low-ranked features, not just top-ranked ones. The novel RIFS2D algorithm integrates multiple RIFS blocks for better feature selection in big data analysis.
Area of Science:
- Biomedical research
- Bioinformatics
- Data science
Background:
- Big data in biomedical research presents challenges in sample recruitment and high-dimensional feature analysis.
- Traditional feature selection methods like incremental feature selection (IFS) focus on top-ranked features, potentially overlooking valuable low-ranked ones.
Purpose of the Study:
- To propose a novel feature selection algorithm, RIFS2D, integrating multiple randomly restarted incremental feature selection (RIFS) blocks.
- To evaluate the effectiveness of RIFS2D compared to existing methods, including IFS and RIFS, in identifying predictive feature subsets.
- To investigate the potential of low-ranked features in achieving robust prediction performance.
Main Methods:
- Development of the RIFS2D algorithm, which combines multiple RIFS blocks for feature selection.
- Comprehensive comparative experiments involving RIFS2D, IFS, RIFS, and other established feature selection algorithms.
- Application of RIFS2D and t-tests for early-stage breast cancer detection, comparing prediction accuracy and drug targeting.
Main Results:
- The study demonstrated that subsets of low-ranked features can achieve promising prediction performance, challenging the conventional focus on top-ranked features.
- RIFS2D outperformed existing feature selection algorithms in comprehensive comparative experiments.
- For early-stage breast cancer detection, RIFS2D-selected features showed superior prediction accuracy and greater drug-targeting potential compared to t-test selected features.
Conclusions:
- Feature selection strategies should consider the predictive power of low-ranked features, as they can yield effective prediction models.
- The RIFS2D algorithm offers a promising approach for feature selection in high-dimensional biomedical data.
- RIFS2D demonstrates potential clinical utility, as evidenced by its performance in early-stage breast cancer detection.
