A Robust Supervised Variable Selection for Noisy High-Dimensional Data.

Area of Science:

• Statistics
• Machine Learning
• Bioinformatics

Background:

• High-dimensional data analysis requires effective dimensionality reduction.
• Existing Minimum Redundancy Maximum Relevance (MRMR) methods can be sensitive to outliers and inefficient.
• Supervised variable selection is crucial for classification tasks.

Purpose of the Study:

• To propose a novel robust method for variable selection in noisy, high-dimensional data.
• To address the limitations of traditional MRMR approaches regarding outlier sensitivity and inefficiency.
• To develop a method suitable for data observed in two groups.

Main Methods:

• Introduced Minimum Regularized Redundancy Maximum Robust Relevance (MRRMRR).
• Employed regularization and robust statistics principles.
• Utilized a regularized coefficient of multiple correlation for redundancy.
• Employed a robust correlation coefficient based on least weighted squares regression for relevance.

Main Results:

• The proposed MRRMRR method demonstrates robustness against outliers and noise.
• Experimental results on real and artificially contaminated datasets validate the method's performance.
• MRRMRR outperforms other dimensionality reduction methods in the presence of noise.

Conclusions:

• MRRMRR is a robust and efficient method for variable selection in high-dimensional, noisy datasets.
• The method offers an improvement over existing MRMR approaches, particularly in challenging data conditions.
• MRRMRR is well-suited for applications in fields dealing with complex, real-world data.