Home // ALLDATA 2019, The Fifth International Conference on Big Data, Small Data, Linked Data and Open Data // View article

A Feature Extraction Framework for Time Series Analysis

Authors:
Angelo Martone
Gaetano Zazzaro
Luigi Pavone

Keywords: data mining; time series analysis; feature extraction; sliding window; similarity measures; pre-processing

Abstract:
With the raise of smart sensors and of the Internet of Things paradigm, there is an increasing demand for performing Data Mining tasks (classification, clustering, outlier detection, etc.) on data stream produced by these inter-connected devices. In particular, Data Mining for time series has gained a relevant importance in the last decade. For these temporal data, feature extraction can be performed using various algorithms and decomposition techniques for time series analysis. In addition, features can also be obtained by sequence comparison techniques, such as dynamic time warping or other measures of similarity. For these reasons, we have designed and implemented a multipurpose and extendable tool for window-based feature extraction from time series data. This paper describes the architecture of the designed tool, named Training Builder, and the current version of its multi-language implementation, which focuses on time series feature extraction, parametric windowing task and data pre-processing. The framework has been applied in the neurological domain where very good results have been achieved for epileptic seizures detection; the case study shows how the Training Builder tool may be very helpful for the next Data Mining tasks.

Pages: 5 to 13

Copyright: Copyright (c) IARIA, 2019

Publication date: March 24, 2019

Published in: conference

ISSN: 2519-8386

ISBN: 978-1-61208-700-9

Location: Valencia, Spain

Dates: from March 24, 2019 to March 28, 2019