Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Survey on Recent Advances in Self-Organizing Maps

Axel Guérin, Pierre Chauvet, Frédéric Saubion

TL;DR

Self-Organising Maps project a $p$-dimensional input space onto a two-dimensional map and organize similar observations into topologically ordered regions. The survey catalogs advances in data management, topology/metrics, learning strategies, visualization, performance, and hyperparameterisation from 2014 onward. Notable contributions include SS-SOM for semi-supervised clustering, FWSOM for feature weighting, DBSOM with Wasserstein distances, non-Euclidean and multilinear distance frameworks, and PSOM/DPSOM for probabilistic clustering and time-series. The review highlights commercial relevance, especially for customer analytics, while noting ongoing challenges in handling diverse data types, missing values, scalability, and automated hyperparameter tuning.

Abstract

Self-organising maps are a powerful tool for cluster analysis in a wide range of data contexts. From the pioneer work of Kohonen, many variants and improvements have been proposed. This review focuses on the last decade, in order to provide an overview of the main evolution of the seminal SOM algorithm as well as of the methodological developments that have been achieved in order to better fit to various application contexts and users' requirements. We also highlight a specific and important application field that is related to commercial use of SOM, which involves specific data management.

A Survey on Recent Advances in Self-Organizing Maps

TL;DR

Self-Organising Maps project a -dimensional input space onto a two-dimensional map and organize similar observations into topologically ordered regions. The survey catalogs advances in data management, topology/metrics, learning strategies, visualization, performance, and hyperparameterisation from 2014 onward. Notable contributions include SS-SOM for semi-supervised clustering, FWSOM for feature weighting, DBSOM with Wasserstein distances, non-Euclidean and multilinear distance frameworks, and PSOM/DPSOM for probabilistic clustering and time-series. The review highlights commercial relevance, especially for customer analytics, while noting ongoing challenges in handling diverse data types, missing values, scalability, and automated hyperparameter tuning.

Abstract

Self-organising maps are a powerful tool for cluster analysis in a wide range of data contexts. From the pioneer work of Kohonen, many variants and improvements have been proposed. This review focuses on the last decade, in order to provide an overview of the main evolution of the seminal SOM algorithm as well as of the methodological developments that have been achieved in order to better fit to various application contexts and users' requirements. We also highlight a specific and important application field that is related to commercial use of SOM, which involves specific data management.
Paper Structure (26 sections, 3 figures, 1 table, 1 algorithm)

This paper contains 26 sections, 3 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. The SOM Algorithm in Brief
  3. Evolution of the SOM Algorithms During the Last Decade
  4. Data Management
  5. Semi-supervized Datasets
  6. Disturbed Datasets
  7. Categorical Data
  8. Particular Data Distributions
  9. Large Datasets
  10. Time and Patterns
  11. Topology and Metrics
  12. Topology
  13. Distances
  14. Neighborhood Relationships
  15. Learning Techniques
  16. ...and 11 more sections

Figures (3)

  • Figure 1: Map of hexagons representing 1 variable from a dataset of 417 entries
  • Figure 2: Evolution of the number of references per year for the ‘Self Organizing Map’ search on the DBLP database (https://dblp.org/)
  • Figure 3: Organization of the survey according to the different components and key features of SOM algorithms