Poenostavitev dreves hierarhičnega gručenja

diplomsko delo

Rok Filipovič (Author), Blaž Zupan (Mentor)

Abstract

Rezanje dreves hierarhičnega gručenja je pomemben proces, vendar je zelo težko oceniti, kje smemo rezati, da izbrana gruča res predstavlja povezavo med svojimi predstavniki. Algoritem, ki nam pomaga to doseči, je pvclust. Za generiranje vzorcev uporablja metodo stremena, ti vzorci pa se nato uporabijo za izračun korelacijskega koeficienta med pari posameznih atributov. Koeficienti se uporabijo kot mera, ki pomaga določiti razdalje med atributi, ki so ključne za delovanje hierarhičnega gručenja. Med iteracijami algoritem primerja gruče in skuša ugotoviti, katere gruče najverjetneje predstavljajo dejanske povezave med atributi. Vendar pa je ena od težav algoritma ta, da za svoje delovanje zahteva veliko časa. Zato v tej nalogi predstavimo alternativo, ki bi dosegla podobne rezultate, vendar bi zahtevala veliko manj časa. Kot kažejo rezultati, nam je z metodo silhuet uspelo izpolniti željen cilj.

Keywords

hierarhično gručenje;dendrogrami;pvclust;metoda stremena;interdisciplinarni študij;univerzitetni študij;diplomske naloge;

Data

Language:	Slovenian
Year of publishing:	2023
Typology:	2.11 - Undergraduate Thesis
Organization:	UL FRI - Faculty of Computer and Information Science
Publisher:	[R. Filipovič]
UDC:	004(043.2)
COBISS:	169193731
Views:	116
Downloads:	6
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Simplification of hierarchical clustering trees
Secondary abstract:	Tree cutting is an important aspect of hierarchical clustering, however, de- termining where to cut often poses a problem, as we would like the clusters to actually represent connections between the objects. An algorithm that helps us achieve this is pvclust. It generates samples through the bootstrap method, which are then used to calculate the correlation between pairs of in- dividual attributes. These values serve as a measure to determine distances that are crucial in hierarchical clustering. Throughout all iterations, the al- gorithm compares which clusters are likely to represent actual connections between features. The only issue is that the algorithm requires a signifi- cant amount of time to operate. Therefore, in this study, we are exploring an alternative that could yield similar results while significantly reducing the required time. Fortunately, it seems that we were able to reproduce sufficient results using the silhouette method.
Secondary keywords:	hierarchical clustering;dendrograms;pvclust;bootstrap method;computer science;computer and information science;computer science and mathematics;interdisciplinary studies;diploma;Računalništvo;Univerzitetna in visokošolska dela;
Type (COBISS):	Bachelor thesis/paper
Study programme:	1000407
Embargo end date (OpenAIRE):	1970-01-01
Thesis comment:	Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages:	48 str.
ID:	19945584