Abstract

There is an abundance of time series data in many domains. Analyzing this data effectively requires deep expertise acquired over many years of practice. Our goal is to develop automated systems for time series analysis that can take advantage of proven methods that yield the best results. Our work is motivated by paleogeosciences time series analysis where the datasets are very challenging and require sophisticated methods to find and quantify subtle patterns. We describe our initial implementation of AutoTS, an automated system for time series analysis that uses semantic workflows to represent sophisticated methods and their constraints. AutoTS extends the WINGS workflow system with new capabilities to customize general methods to specific datasets based on key characteristics of the data. We discuss general methods for spectral analysis and their implementation in AutoTS.

Metadata

publication

Proceedings of the Sixth Workshop on Mining and Learning from Time Series …, 2020

year

2020

publication date

2020

authors

Deborah Khider, Pratheek Athreya, Varun Ratnakar, Yolanda Gil, Feng Zhu, Myron Kwan, Julien Emile-Geay

link

https://kdd-milets.github.io/milets2020/papers/MiLeTS2020_paper_15.pdf

resource_link

https://kdd-milets.github.io/milets2020/papers/MiLeTS2020_paper_15.pdf

journal

Proceedings of the Sixth Workshop on Mining and Learning from Time Series (MiLeTS), held in conjunction with the 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD'20)