A general framework for never-ending learning from time series streams

Chen, Yanping; Hao, Yuan; Rakthanmanon, Thanawin; Zakaria, Jesin; Hu, Bing; Keogh, Eamonn

doi:10.1007/s10618-014-0388-4

Cited by 31 publications

(21 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…[3]), there is almost no research dedicated to time series labeling. As far as we are aware, there are four papers on time series labeling [1,9,12,13].…”

Section: Related Workmentioning

confidence: 99%

“…In this paper we introduce the Robust Time series Labeling (RTL) algorithm, that aims to minimize the labeling effort without affecting the overall quality of TSC tasks. The contribution of the paper is fourfold: (1) through an extensive review of an existing time series labeling algorithm we identify the key features needed to have a robust method for reducing the labeling effort while maintaining high quality labels, (2) we present the RTL algorithm that is based on these key features, (3) as part of the RTL algorithm we introduce a novel zoom-in step that secures the quality of the assigned labels, and (4) we show RTL's ability to increase the efficiency and robustness of the labeling procedure across a wide variety of time series datasets. RTL clears the path for TSC tasks in many application domains where large amounts of complex data are generated and need to be labeled continuously.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

RTL: A Robust Time Series Labeling Algorithm

Leeuwen¹,

Bosma²,

Born³

et al. 2021

Advances in Intelligent Data Analysis XIX

View full text Add to dashboard Cite

Time series classification is one of the most important problems in data mining. With the growth in availability of time series data, many novel classification algorithms have been proposed. Despite the promising progress in accuracy, the performance of many algorithms still strongly depends on an initial training session containing labeled examples of all classes to be learned. In most realistic applications, however, labels are lacking or only partially available; limiting the practical applicability of time series classification algorithms with this requirement. To remedy this, we introduce the Robust Time series Labeling (RTL) algorithm and show its ability to increase labeling accuracy and robustness across a wide variety of time series datasets. Given its flexibility, the RTL algorithm can successfully be applied in many real-life situations.

show abstract

“…[3]), there is almost no research dedicated to time series labeling. As far as we are aware, there are four papers on time series labeling [1,9,12,13].…”

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

RTL: A Robust Time Series Labeling Algorithm

Leeuwen¹,

Bosma²,

Born³

et al. 2021

Advances in Intelligent Data Analysis XIX

View full text Add to dashboard Cite

show abstract

“…34 The negative class consists of three abnormal types**: (A1) R-on-T PVC, (A2) supraventricular premature or ectopic beat, and (A3) PVC. 36 Because we seek to discover shapelets that capture specific abnormal types, LTSpAUC uses optional constraints (4.16) by setting K À = K = 40 so that shapelets tend to match abnormal time series, unlike in the Experimental Setup section. Figure 9a shows the ROC curves from test data for LTSpAUC and LTSfAUC.…”

Section: Premature Ventricular Contraction Detection In Medicinementioning

confidence: 99%

LTSpAUC: Learning Time-Series Shapelets for Partial AUC Maximization

et al. 2020

View full text Add to dashboard Cite

Shapelets are discriminative segments used to classify time-series instances. Shapelet methods that jointly learn both classifiers and shapelets have been studied in recent years because such methods provide both interpretable results and superior accuracy. The partial area under the receiver operating characteristic curve (pAUC) for a low range of false-positive rates (FPR) is an important performance measure for practical cases in industries such as medicine, manufacturing, and maintenance. In this article, we propose a method that jointly learns both shapelets and a classifier for pAUC optimization in any FPR range, including the full AUC. In addition, we propose the following two extensions for shapelet methods: (1) reducing algorithmic complexity in time-series length to linear time and (2) explicitly determining the classes that shapelets tend to match. Comparing with state-of-theart learning-based shapelet methods, we demonstrated the superiority of pAUC on UCR time-series data sets and its effectiveness in industrial case studies from medicine, manufacturing, and maintenance.

show abstract

“…Research on the early classification of real number sequences and feature extraction interpretability was proposed in [357]- [360]. A general framework for the study of continuous learning of real number sequences was introduced in [361]. The minimum description length (MDL) and piecewise aggregate approximation (PAA) sequence of real number were used to reduce the dimension in [362], [363].…”

Section: ) Mobile Terminal Behavior Analysis Of Network Usersmentioning

confidence: 99%

Behavior Analysis for Electronic Commerce Trading Systems: A Survey

et al. 2019

View full text Add to dashboard Cite

With the rapid development of the Internet and the continuous progress of computer science and technology, Internet-based online trading has developed rapidly and improved the convenience of everyday life. Increasingly, people conduct business activities through online transactions and payment methods, and the development prospects for online transactions are quite broad. In addition, e-commerce trading systems have attracted considerable attention. Many studies have been conducted to analyze the behaviors of these systems. This paper focuses on analyzing both system behavior and user behavior in e-commerce trading systems. To reflect the characteristics of online trading systems, system behavior reflects the strength of the system, while user behavior reflects various user actions. Consistency, which is a basis for determining whether two objects are similar, provides a reference for designers and modelers. This paper also focuses on analyzing the consistency of system behavior, especially regarding the measurement and diagnosis of e-commerce trading systems. Then, we study the existing user behavior analysis methods for e-commerce trading systems. We divide user behaviors into four groups: web browsing behavior, keystroke behavior, network transaction behavior, and mobile terminal behavior. The performance and correlation analysis techniques of these methods are discussed and compared in detail. Finally, we provide ideas for future research involving behavior analysis for online trading systems.

show abstract

A general framework for never-ending learning from time series streams

Cited by 31 publications

References 46 publications

RTL: A Robust Time Series Labeling Algorithm

RTL: A Robust Time Series Labeling Algorithm

LTSpAUC: Learning Time-Series Shapelets for Partial AUC Maximization

Behavior Analysis for Electronic Commerce Trading Systems: A Survey

Contact Info

Product

Resources

About