STLBRF: an improved random forest algorithm based on standardized-threshold for feature screening of gene expression...

Area of Science:

• Biostatistics
• Bioinformatics
• Machine Learning

Background:

• Traditional random forest (RF) algorithms struggle with noise and parameter interference in feature selection.
• Direct noise removal can introduce significant bias in biostatistical analyses.
• Accurate feature gene selection is crucial for biomarker discovery and expression analysis.

Purpose of the Study:

• To develop an improved random forest algorithm for robust feature gene selection.
• To address the limitations of traditional RF in handling noisy gene expression data.
• To enhance the accuracy and control of feature selection in biostatistical applications.

Main Methods:

• Developed a novel standardized threshold and loops based random forest (STLBRF) algorithm.
• Integrated backward elimination and K-fold cross-validation with a standardized error increment threshold.
• Compared STLBRF against ridge, lasso, elastic net, and traditional RF using three real gene expression datasets.

Main Results:

• STLBRF demonstrated superior effectiveness in feature gene selection compared to existing methods.
• The algorithm provided better control over the number of selected feature genes.
• Validation using a Random Forest classifier confirmed the reliability of STLBRF-selected genes.

Conclusions:

• The STLBRF algorithm offers a robust solution for feature gene selection in the presence of noise.
• This method provides reliable technical support for feature expression analysis and biomarker research.
• STLBRF enhances the accuracy and controllability of gene selection, overcoming traditional RF limitations.