Bayesian Clustering Factor Models.

Area of Science:

• Statistics
• Machine Learning
• Bioinformatics

Background:

• Clustering and dimension reduction are crucial for analyzing complex datasets.
• Existing methods often struggle with simultaneous application or accurate parameter selection.
• Bayesian approaches offer a robust framework for statistical modeling and inference.

Purpose of the Study:

• To introduce a novel Bayesian framework for concomitant dimension reduction and clustering.
• To develop an information criterion for selecting the optimal number of clusters and factors.
• To evaluate the performance of the proposed framework against existing methods.

Main Methods:

• Developed a novel class of Bayesian clustering factor models with Gaussian mixture distributions for common factors.
• Implemented a Gibbs sampler for efficient posterior distribution exploration.
• Proposed an information criterion for model selection (number of clusters and factors).

Main Results:

• The proposed inferential approach effectively quantifies uncertainty.
• Simulation studies demonstrated favorable performance of the information criterion in selecting the correct number of clusters and factors compared to two competitor methods.
• The framework was successfully applied to opioid use disorder recovery data.

Conclusions:

• The novel Bayesian framework provides a powerful tool for simultaneous dimension reduction and clustering.
• The proposed information criterion enhances model selection accuracy.
• This framework has potential applications in personalized healthcare, such as tailoring treatments for opioid use disorder recovery.