Unsupervised clustering analysis: a multiscale complex networks approach

TL;DR

This paper introduces a multiscale complex networks approach for unsupervised clustering, extending modularity-based algorithms to identify data clusters at multiple resolutions, and evaluates their effectiveness on benchmark datasets.

Contribution

It proposes extending multiresolution modularity algorithms for data clustering, providing a novel multiscale analysis method within the complex networks framework.

Findings

Algorithms successfully identify clusters at multiple scales

Performance comparable to existing clustering methods on benchmarks

Demonstrates the effectiveness of complex networks in unsupervised clustering

Abstract

Unsupervised clustering, also known as natural clustering, stands for the classification of data according to their similarities. Here we study this problem from the perspective of complex networks. Mapping the description of data similarities to graphs, we propose to extend two multiresolution modularity based algorithms to the finding of modules (clusters) in general data sets producing a multiscales' solution. We show the performance of these reported algorithms to the classification of a standard benchmark of data clustering and compare their performance.