Model-based hierarchical clustering algorithm using Bayesian networks
Author
Jurgelenaite, Rasa
Term
10. Term
Education
Publication year
2003
Abstract
Denne afhandling præsenterer og afprøver en modelbaseret klyngealgoritme. Klyngedannelse handler om at gruppere lignende observationer uden forudgående etiketter. I vores tilgang beskrives hver klynge med en dirigeret grafisk model (et Bayesiansk netværk), som indfanger afhængighederne mellem variablerne. Algoritmen bygger klyngerne trin for trin med en hierarkisk strategi. Til at lære strukturen af netværkene fra data anvendes to kendte metoder: PC-algoritmen og KES-algoritmen. Metoden evalueres på genekspressionsdata (målinger af, hvor aktive mange gener er), og resultaterne viser en fornuftig ydeevne for den foreslåede tilgang.
This thesis presents and tests a model-based clustering algorithm. Clustering groups similar observations without prior labels. In our approach, each cluster is described by a directed graphical model (a Bayesian network) that captures dependencies among variables. The algorithm builds clusters step by step using a hierarchical strategy. To learn the network structure from data, it employs two established methods: the PC algorithm and the KES algorithm. We evaluate the method on gene expression data (measurements of how active many genes are), and the results indicate reasonable performance of the proposed approach.
[This abstract was generated with the help of AI]
Documents