Constrained parsimonious model-based clustering.

Area of Science:

• Statistics
• Data Mining
• Computational Biology

Background:

• Model-based clustering is a powerful technique for data analysis.
• Existing methods often lack flexibility in handling constraints.
• Parsimonious models offer interpretability but can be restrictive.

Purpose of the Study:

• Introduce a novel methodology for constrained parsimonious model-based clustering.
• Provide a flexible framework with a tunable parameter to control constraint strength.
• Bridge the gap between existing parsimonious models for smoother transitions.

Main Methods:

• Developed a constrained optimization framework for model-based clustering.
• Integrated 14 parsimonious models, including normal component cases.
• Proposed new information criteria for parameter selection.

Main Results:

• The methodology allows for a smooth transition among models.
• Mathematically well-defined problems are formulated, preventing spurious solutions.
• Simulation studies and a COVID data application demonstrate effectiveness.

Conclusions:

• The proposed constrained clustering methodology offers enhanced flexibility and robustness.
• Novel information criteria facilitate practical application and parameter tuning.
• The approach is valuable for diverse datasets, including epidemiological data.