Learning accurate and interpretable models based on regularized random forests regression.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Biological research often integrates diverse data sources for comprehensive understanding.
• Extracting key information and interpretable insights from complex biological datasets is crucial.
• Existing methods struggle to balance prediction performance with interpretability, especially for nonlinear biological data.

Purpose of the Study:

• To develop an effective algorithm for biological data analysis that simultaneously extracts decision rules and selects critical features.
• To address the challenge of interpreting nonlinear relationships in biological data while maintaining high prediction accuracy.
• To provide a tool that aids in both prediction and interpretation for complex biological problems.

Main Methods:

• Proposed a novel rule-based regression algorithm utilizing 1-norm regularized random forests.
• The algorithm is designed for continuous target outcomes in biological regression problems.
• Simultaneously extracts a concise set of regression rules and eliminates redundant features.

Main Results:

• The approach successfully constructed a significantly smaller set of regression rules on biological datasets.
• Achieved prediction performance comparable to standard random forests regression.
• Demonstrated effective feature selection, utilizing a subset of relevant attributes.

Conclusions:

• The developed algorithm shows high potential for aiding prediction in biological studies.
• Offers enhanced interpretability of nonlinear relationships within biological data.
• Provides a valuable tool for researchers needing to understand complex biological systems.