Leveraging Rank Information for Robust Regression Analysis: A Nomination Sampling Approach.

Area of Science:

• Statistics
• Machine Learning

Background:

• Traditional mean regression struggles with datasets containing outliers.
• Simple random sampling (SRS) may not yield representative samples in the presence of extensive outliers.

Purpose of the Study:

• Introduce a novel methodology for robust regression analysis.
• Enhance sample representativeness and regression accuracy in the presence of outliers.
• Improve upon traditional mean regression methods.

Main Methods:

• Leverage median nomination sampling (MedNS) using ranking information for training data.
• Propose a new loss function integrating rank information from MedNS data.
• Develop an alternative approach translating MedNS median regression to SRS.

Main Results:

• The proposed MedNS methodology enhances sample representativeness.
• The novel loss function provides a robust regression approach.
• Simulation studies show higher relative efficiency (RE) compared to SRS counterparts.
• The method was applied to a real-world body fat analysis dataset.

Conclusions:

• The novel MedNS methodology offers a robust regression solution superior to traditional SRS methods.
• Integrating rank information improves regression model fitting in the presence of outliers.
• The proposed approach demonstrates practical utility in real-world data analysis.