Mining features for biomedical data using clustering tree ensembles.

Area of Science:

• Biomedical data science
• Machine learning applications
• Bioinformatics

Background:

• Biomedical datasets rapidly increase, posing challenges for machine learning (ML) interpretability and performance.
• Many datasets exhibit limited variance or identical features with differing class labels, hindering ML model effectiveness.

Purpose of the Study:

• To address the problem of low-variance and poor feature discrimination in biomedical data.
• To propose and validate a novel feature induction method to improve data quality for ML.

Main Methods:

• A target-informed feature induction method utilizing tree ensemble learning.
• Validation of the proposed method on multi-target prediction tasks.

Main Results:

• The method introduces greater variance into data representation.
• Demonstrated improvement in discriminating between data instances.
• Enhanced predictive performance of machine learning learners.

Conclusions:

• The proposed feature induction method effectively tackles data quality issues in biomedical datasets.
• This approach holds potential for improving the reliability and accuracy of ML models in biomedical research.