Rotation of random forests for genomic and proteomic classification problems.

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• High-dimensional gene expression data classification presents challenges.
• Random Forests offer robustness for large feature sets in classification.
• Existing ensemble classifiers have limitations in accuracy and robustness.

Purpose of the Study:

• Introduce Rotation Forest as an ensemble of decision trees.
• Evaluate Rotation Forest's classification performance on microarray datasets.
• Assess the potential of Rotation Forest to improve existing ensembles like Random Forest.

Main Methods:

• Utilized decision trees as base classifiers within the Rotation Forest framework.
• Applied Rotation Forest to 14 diverse microarray datasets containing genomic and proteomic data.
• Investigated the performance of a modified Random Forest incorporating rotation techniques.

Main Results:

• Rotation Forest demonstrated superior classification performance across multiple datasets.
• The proposed rotation of Random Forests also showed improved accuracy and robustness.
• Rotation Forest outperformed widely used ensemble classifiers, including standard Random Forests, on the majority of tested datasets.

Conclusions:

• Rotation Forest is a highly effective classification technique for high-dimensional biological data.
• Enhancing Random Forests with rotation improves their accuracy and robustness.
• Rotation Forest offers a promising alternative for complex classification tasks in genomics and proteomics.