MatArcs: An Exploratory Data Analysis of Recurring Patterns in Multivariate Time Series

Abstract

The development of algorithms in exploratory data analysis (EDA) requires handling a large amount of data and presenting it in a user-friendly environment. However, the results of EDA are often not self-explanatory. Therefore, we present MatArcs, a novel visualization technique for the detection of recurring time series patterns. Major advantages of MatArcs are the dimensionality reduction of complex multivariate data and the localization of recurring patterns. Geared to the needs of various users which have deviating objectives in finding patterns, we present an algorithm which is able to fullfil the multilayered task of exploratory data analysis; i.e., i) processing multivariate and large time series (even real and noisy data may not pose a problem), ii) the versatility to be applied in fields of theoretical research, iii) extracting the main characteristics of a data set by using significant graphical attributes and iv) identified patterns can be synthesized to new patterns. With regards to the field of visual analytics our proposed approach provides a visualization which extends/increases the retentiveness of the user to get the targeted characteristics, even in multivariate time series.

@inproceedings{MatArcs,
author = {Gaebler, Julia and Spiegel, Stephan and Albayrak, Sahin},
title = {MatArcs: An Exploratory Data Analysis of Recurring Patterns in Multivariate Time Series},
booktitle = {Lecture Notes in Artificial Intelligence (LNAI) / ECML-PKDD 2012: Proceedings of Workshop on New Frontiers in Mining Complex Patterns (NFMCP)},
year = {2012},
location = {Bristol, UK},
publisher = {Springer},
}
Autoren:
Julia Gäbler, Stephan Spiegel, Sahin Albayrak
Kategorie:
Tagungsbeitrag
Jahr:
2012
Ort:
Lecture Notes in Artificial Intelligence (LNAI) / ECML-PKDD 2012: Proceedings of Workshop on New Frontiers in Mining Complex Patterns (NFMCP), Bristol, UK