Improved initialisation of model-based clustering using Gaussian\n hierarchical partitions

2015/07/25 by Luca Scrucca, Adrian E. Raftery, Scrucca, Luca +1
Biochemistry, Genetics and Molecular Biology · Computer Science · #Advanced Clustering Algorithms Research #Bayesian Methods and Mixture Models #Data Management and Algorithms #FOS: Computer and information sciences #Gene expression and cancer classification #Methodology (stat.ME)

paper · pdf · doi:10.48550/arxiv.1507.07114

openalex publication_date 2015/07/25 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Initialisation of the EM algorithm in model-based clustering is often\ncrucial. Various starting points in the parameter space often lead to different\nlocal maxima of the likelihood function and, so to different clustering\npartitions. Among the several approaches available in the literature,\nmodel-based agglomerative hierarchical clustering is used to provide initial\npartitions in the popular MCLUST R package. This choice is computationally\nconvenient and often yields good clustering partitions. However, in certain\ncircumstances, poor initial partitions may cause the EM algorithm to converge\nto a local maximum of the likelihood function. We propose several simple and\nfast refinements based on data transformations and illustrate them through data\nexamples.\n

Related

• An Experimental Comparison of Several Clustering and Initialization Methods
• A flexible EM-like clustering algorithm for noisy data
• A robust model-based clustering based on the geometric median and the Median Covariation Matrix
• An Evolutionary Algorithm with Crossover and Mutation for Model-Based\n Clustering
• Bayesian Hierarchical Clustering with Exponential Family: Small-Variance Asymptotics and Reducibility