Paper Type
Short
Paper Number
1149
Description
Time series classification has become one of the most important tasks in machine learning applications. It is a supervised learning task to classify classes of time series. This paper proposes a novel feature-based time series classification that uses estimated autoregressive fractionally integrated moving average (ARFIMA) parameters for each time series as a feature-based representation. This research makes an innovative contribution to the feature-based time series classification literature by employing ARFIMA coefficients to characterize time series patterns that have not been previously investigated. This technique is able to derive a smaller set of parameters from the time series model than the length of its series, process sets of time series exhibiting long memory, and process those time series with different lengths. Well-known benchmark datasets were used to evaluate the proposed method, and we obtained competitive performance in comparison with those provided by the state-of-the-art methods.
Recommended Citation
Sumara, Rauzan; Homenda, Wladyslaw; and Pedrycz, Witold, "ARFIMA for Feature-Based Time Series Classification" (2024). PACIS 2024 Proceedings. 2.
https://aisel.aisnet.org/pacis2024/track03_ba/track03_ba/2
ARFIMA for Feature-Based Time Series Classification
Time series classification has become one of the most important tasks in machine learning applications. It is a supervised learning task to classify classes of time series. This paper proposes a novel feature-based time series classification that uses estimated autoregressive fractionally integrated moving average (ARFIMA) parameters for each time series as a feature-based representation. This research makes an innovative contribution to the feature-based time series classification literature by employing ARFIMA coefficients to characterize time series patterns that have not been previously investigated. This technique is able to derive a smaller set of parameters from the time series model than the length of its series, process sets of time series exhibiting long memory, and process those time series with different lengths. Well-known benchmark datasets were used to evaluate the proposed method, and we obtained competitive performance in comparison with those provided by the state-of-the-art methods.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
Analytics