Clustering of longitudinal interval-valued data via mixture distribution under covariance separability

Area of Science:

• Statistics
• Data Science
• Biostatistics

Background:

• Clustering interval-valued data presents unique challenges.
• Traditional methods may struggle with the complexity of repeated measurements.
• Model-based clustering (M-clustering) offers a potential solution.

Purpose of the Study:

• To develop and evaluate a novel M-clustering approach for interval-valued data.
• To investigate the benefits of using a separable covariance matrix structure.
• To apply the method to real-world longitudinal health data.

Main Methods:

• Treating interval-valued data as matrix variate data.
• Assuming a separable covariance matrix for M-clustering.
• Comparing performance against other covariance structures via numerical studies.
• Applying the method to longitudinal blood pressure data from the NGHS.

Main Results:

• The separable covariance matrix structure requires fewer samples for valid clustering.
• This structure enhances the accuracy in determining the correct number of clusters.
• The M-clustering method effectively clusters longitudinal blood pressure data.

Conclusions:

• The proposed M-clustering with a separable covariance matrix is efficient and accurate for interval-valued data.
• This approach offers advantages over traditional methods, especially for longitudinal studies.
• The method demonstrates practical utility in analyzing complex health datasets.