A unified framework for model-based clustering, linear regression and multiple cluster structure detection

TL;DR

This paper introduces a comprehensive framework that unifies model-based clustering, linear regression, and detection of multiple cluster structures, enhancing analysis capabilities for complex data.

Contribution

It presents a novel unified approach that combines Gaussian mixture models with variable selection and multiple clustering detection, including a new identifiability condition.

Findings

Effective in simulated datasets

Successfully applied to real datasets

Detects multiple cluster structures

Abstract

A general framework for dealing with both linear regression and clustering problems is described. It includes Gaussian clusterwise linear regression analysis with random covariates and cluster analysis via Gaussian mixture models with variable selection. It also admits a novel approach for detecting multiple clusterings from possibly correlated sub-vectors of variables, based on a model defined as the product of conditionally independent Gaussian mixture models. A necessary condition for the identifiability of such a model is provided. The usefulness and effectiveness of the described methodology are illustrated using simulated and real datasets.