Measuring similarity of univariate time series

undergraduate thesis

Maša Kljun (Author), Erik Štrumbelj (Mentor)

Abstract

In this thesis we review 12 time series similarity measures and 3 classifications of these measures into groups. We view similarity measures in terms of time complexity, support of time series of different lengths, and normalization. With empirical evaluation we check measures' invariances to warping and scaling, their clustering performance, and how similar they are. We find out that although several measures perform well on average no measure performs well in all cases. We see that the Piccolo distance is invariant to warping and scaling, and that it stands out with its clustering performance and linear time complexity. We also see that compression-based measures perform poorly on average.

Keywords

time series;similarity measures;classification of similarity measures;clustering;computer science;computer and information science;computer science and mathematics;interdisciplinary studies;diploma;

Data

Language:	English
Year of publishing:	2020
Typology:	2.11 - Undergraduate Thesis
Organization:	UL FRI - Faculty of Computer and Information Science
Publisher:	[M. Kljun]
UDC:	004(043.2)
COBISS:	28472067
Views:	850
Downloads:	229
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	Slovenian
Secondary title:	Merjenje podobnosti univariatnih časovnih vrst
Secondary abstract:	V diplomskem delu obravnavamo 12 mer podobnosti za časovne vrste in 3 delitve le-teh v skupine. Mere podobnosti obravnavamo z vidika njihovih časovnih zahtevnosti ter drugih lastnosti, kot so sposobnost primerjave časovnih vrst različnih dolžin in normalizacija razdalje. Empirično preverimo invariantnost mer na ukrivljanje in množenje s skalarjem, njihovo uspešnost pri gručenju in kako podobne so si. Ugotovimo, da nobena mera ni ustrezna v vseh primerih, saj ima vsaka svoje pomanjkljivosti. Vidimo, da je razdalja Piccolo invariantna na ukrivljanje in množenje s skalarjem ter da izstopa s svojo linearno časovno zahtevnostjo in dobrim rezultatom pri gručenju. Vidimo tudi, da mere, ki temeljijo na kompresiji, v povprečju ne dajejo dobrih rezultatov.
Secondary keywords:	časovne vrste;mere podobnosti;klasifikacija mer podobnosti;gručenje;računalništvo;računalništvo in informatika;računalništvo in matematika;interdisciplinarni študij;univerzitetni študij;diplomske naloge;
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:	42 str.
ID:	12027632