Korelacijsko grupiranje vremenskih nizova

undergraduate thesis

• Title (english): Correlation Based Clustering of Time Series
• Author: Matea Torbarina
• Mentor: Zvonko Kostanjčar (mentor)
• Committee member: Zvonko Kostanjčar (predsjednik povjerenstva)
• Committee member: Mate Puljiz (član povjerenstva)
• Committee member: Stjepan Begušić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2017-09-12, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering

Abstract

Problem primjene tradicionalnih algoritama grupiranja na visoko-dimenzionalnim vremenskim nizovima je unaprijed definirani broj grupa. Veliki izazov predstavlja upravo česta vremenska promjenjivost grupiranja i unaprijed nepoznat broj grupa. U modeliranju sličnosti parova vremenskih nizova kao korisna statistička mjera koristi se mjera korelacije, na kojoj se zasnivaju novi algoritmi grupiranja. U radu se istražuju algoritmi korelacijskog grupiranja, te se razvija novi algoritam za primjenu na vremenskim nizovima. Istražene su razne inačice, svojstva i metode estimacije korelacijskih matrica vremenskih nizova, nakon čega je razvijen algoritam korelacijskog grupiranja zasnovan na particioniranju. Razvijeni algoritam je ispitan na financijskim podacima danim za kompanije iz 9 različitih sektora.

Abstract (english)

The problem of application of traditional clustering algorithm on high-dimensional time series is pre defined cluster number. Frequent clustering fluctuations is another big challenge in this kind of statistical analysis that experts aspire to solve. In modelling the similarity of time series pairs, there is very useful method of grouping, based on correlation as a measure of distance. This paper reviews the development of an correlation clustering and algorithm for time series application. The various variants, properties and methods of estimating correlation matrices were researched, and a partitioning based correlation algorithm was developed. The developed algorithm were tested on a financial data sets of a 9 companies, where each of them belongs to a different sector.

• Keywords: korelacijsko grupiranje; k-means; vremenske serije; mjere udaljenosti
• Keywords (english): correlation clustering; k-means; time series; measure of distance
• Language: croatian
• URN:NBN: urn:nbn:hr:168:786667
• Study programme: Title: Electrical Engineering and Information Technology; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka elektrotehnike i informacijske tehnologije (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka elektrotehnike i informacijske tehnologije)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2019-03-14 01:39:15