Unsupervised Liu-type shrinkage estimators for mixture of regression models

Area of Science:

• Statistics
• Machine Learning
• Biostatistics

Background:

• Probabilistic regression models are key for population heterogeneity analysis.
• Multicollinearity among covariates can lead to unreliable estimates.
• Existing methods struggle with multicollinearity challenges.

Purpose of the Study:

• To develop novel Liu-type shrinkage methods for robust coefficient estimation.
• To address multicollinearity in probabilistic regression models using unsupervised learning.
• To enhance the accuracy of population heterogeneity analysis.

Main Methods:

• Developed Liu-type shrinkage methods.
• Employed an unsupervised learning approach.
• Utilized classification and stochastic expectation-maximization algorithms for evaluation.

Main Results:

• Proposed methods demonstrated superior performance over Ridge and maximum likelihood.
• Numerical simulations confirmed the effectiveness of the new techniques.
• Successful application to bone mineral data analysis in older women.

Conclusions:

• Liu-type shrinkage methods offer a robust solution for multicollinearity.
• The proposed unsupervised learning approach enhances model reliability.
• These methods have practical implications for health data analysis, including bone density studies.