Litcius/Paper detail

ClusterValidityIndices.jl: Batch and IncrementalMetrics for Unsupervised Learning

Sasha Petrenko, Donald C. Wunsch

2022The Journal of Open Source Software11 citationsDOIOpen Access PDF

Abstract

ClusterValidityIndices.jl is a Julia package for evaluating the performance of clustering algorithms without the aid of supervised labels. Cluster Validity Indices (CVI) provide a metric of the over-or under-partitioning of an arbitrary clustering algorithm with only the original data and labels assigned by the clustering algorithm. Furthermore, there exist formulations of every CVI such that they may run incrementally (i.e. Incremental CVIs, or ICVI), streaming alongside the clustering algorithm and producing the same results as in their batch implementations. Using a standard interface, each CVI in this package can be run with any clustering algorithm to produce a metric of that algorithm's performance in scenarios where explicit supervised labels do not exist, which is extremely useful in real-world applications where that is often the case.

Topics & Concepts

Cluster analysisComputer scienceMetric (unit)Data miningImplementationCanopy clustering algorithmCorrelation clusteringCURE data clustering algorithmData stream clusteringInterface (matter)Artificial intelligenceAlgorithmMachine learningParallel computingProgramming languageEconomicsBubbleMaximum bubble pressure methodOperations managementAdvanced Clustering Algorithms ResearchComplex Network Analysis TechniquesData Stream Mining Techniques
ClusterValidityIndices.jl: Batch and IncrementalMetrics for Unsupervised Learning | Litcius