Co-clustering of Time-Dependent Data via the Shape Invariant Model

Area of Science:

• Statistics
• Data Science
• Machine Learning

Background:

• Multivariate time-dependent data are common across many fields.
• Existing methods struggle to model subject heterogeneity and relationships between time and variables simultaneously.
• There's a need for robust techniques to analyze complex functional and longitudinal data.

Purpose of the Study:

• To propose a new co-clustering methodology for simultaneous grouping of individuals and variables.
• To handle both functional and longitudinal data structures.
• To provide parsimonious and interpretable summaries of complex time-dependent data.

Main Methods:

• Embedding the shape invariant model within the latent block model.
• Utilizing a modified SEM-Gibbs algorithm for estimation.
• Incorporating concepts from the curve registration framework.
• Allowing user-defined cluster specifications.

Main Results:

• The proposed method effectively partitions data matrices into homogeneous blocks.
• It explicitly models time evolution and subject heterogeneity.
• The procedure yields interpretable clusters for complex datasets.

Conclusions:

• The novel co-clustering approach offers a powerful tool for analyzing multivariate time-dependent data.
• It facilitates the understanding of functional and longitudinal data by identifying homogeneous subgroups.
• The method provides parsimonious summaries beneficial even in low-dimensional settings.